Wednesday Dec 08, 2010

Sun Blade X6275 M2 Cluster with Sun Storage 7410 Performance Running Seismic Processing Reverse Time Migration

This Oil & Gas benchmark highlights both the computational performance improvements of the Sun Blade X6275 M2 server module over the previous genernation server and the linear scalability achievable for the total application throughput using a Sun Storage 7410 system to deliver almost 2 GB/sec I/O effective write performance.

Oracle's Sun Storage 7410 system attached via 10 Gigabit Ethernet to a cluster of Oracle's Sun Blade X6275 M2 server modules was used to demonstrate the performance of a 3D VTI Reverse Time Migration application, a heavily used geophysical imaging and modeling application for Oil & Gas Exploration. The total application throughput scaling and computational kernel performance improvements are presented for imaging two production sized grids using 800 input samples.

  • The Sun Blade X6275 M2 server module showed up to a 40% performance improvement over the previous generation server module with super-linear scalability to 16 nodes for the 9-Point Stencil used in this Reverse Time Migration computational kernel.

  • The balanced combination of Oracle's Sun Storage 7410 system over 10 GbE to the Sun Blade X6275 M2 server module cluster showed linear scalability for the total application throughput, including the I/O and MPI communication, to produce a final 3-D seismic depth imaged cube for interpretation.

  • The final image write time from the Sun Blade X6275 M2 server module nodes to Oracle's Sun Storage 7410 system achieved 10GbE line speed of 1.25 GBytes/second or better write performance. The effects of I/O buffer caching on the Sun Blade X6275 M2 server module nodes and 34 GByte write optimized cache on the Sun Storage 7410 system gave up to 1.8 GBytes/second effective write performance.

Performance Landscape

Server Generational Performance Improvements

Performance improvements for the Reverse Time Migration computational kernel using a Sun Blade X6275 M2 cluster are compared to the previous generation Sun Blade X6275 cluster. Hyper-threading was enabled for both configurations allowing 24 OpenMP threads for the Sun Blade X6275 M2 server module nodes and 16 for the Sun Blade X6275 server module nodes.

Sun Blade X6275 M2 Performance Improvements
Number Nodes Grid Size - 1243 x 1151 x 1231 Grid Size - 2486 x 1151 x1231
X6275 Kernel Time (sec) X6275 M2 Kernel Time (sec) X6275 M2 Speedup X6275 Kernel Time (sec) X6275 M2 Kernel Time (sec) X6275 M2 Speedup
16 306 242 1.3 728 576 1.3
14 355 271 1.3 814 679 1.2
12 435 346 1.3 945 797 1.2
10 541 390 1.4 1156 890 1.3
8 726 555 1.3 1511 1193 1.3

Application Scaling

Performance and scaling results of the total application, including I/O, for the reverse time migration demonstration application are presented. Results were obtained using a Sun Blade X6275 M2 server cluster with a Sun Storage 7410 system for the file server. The servers were running with hyperthreading enabled, allowing for 24 OpenMP threads per server node.

Application Scaling Across Multiple Nodes
Number Nodes Grid Size - 1243 x 1151 x 1231 Grid Size - 2486 x 1151 x1231
Total Time (sec) Kernel Time (sec) Total Speedup Kernel Speedup Total Time (sec) Kernel Time (sec) Total Speedup Kernel Speedup
16 501 242 2.1\* 2.3\* 1060 576 2.0 2.1\*
14 583 271 1.8 2.0 1219 679 1.7 1.8
12 681 346 1.6 1.6 1420 797 1.5 1.5
10 807 390 1.3 1.4 1688 890 1.2 1.3
8 1058 555 1.0 1.0 2085 1193 1.0 1.0

\* Super-linear scaling due to the compute kernel fitting better into available cache for larger node counts

Image File Effective Write Performance

The performance for writing the final 3D image from a Sun Blade X6275 M2 server cluster over 10 Gigabit Ethernet to a Sun Storage 7410 system are presented. Each server allocated one core per node for MPI I/O thus allowing 22 OpenMP compute threads per node with hyperthreading enabled. Captured performance analytics from the Sun Storage 7410 system indicate effective use of its 34 Gigabyte write optimized cache.

Image File Effective Write Performance
Number Nodes Grid Size - 1243 x 1151 x 1231 Grid Size - 2486 x 1151 x1231
Write Time (sec) Write Performance (GB/sec) Write Time (sec) Write Performance (GB/sec)
16 4.8 1.5 10.2 1.4
14 5.0 1.4 10.2 1.4
12 4.0 1.8 11.3 1.3
10 4.3 1.6 9.1 1.6
8 4.6 1.5 9.7 1.5

Note: Performance results better than 1.3GB/sec related to I/O buffer caching on server nodes.

Configuration Summary

Hardware Configuration:

8 x 2 node Sun Blade X6275 M2 server nodes, each node with
2 x 2.93 GHz Intel Xeon X5670 processors
48 GB memory (12 x 4 GB at 1333 MHz)
1 x QDR InfiniBand Host Channel Adapter

Sun Datacenter InfiniBand Switch IB-36
Sun Network 10 GbE Switch 72p

Sun Storage 7410 system connected via 10 Gigabit Ethernet
4 x 17 GB STEC ZeusIOPs SSD mirrored - 34 GB
40 x 750 GB 7500 RPM Seagate SATA disks mirrored - 14.4 TB
No L2ARC Readzilla Cache

Software Configuration:

Oracle Enterprise Linux Server release 5.5
Oracle Message Passing Toolkit 8.2.1c (for MPI)
Oracle Solaris Studio 12.2 C++, Fortran, OpenMP

Benchmark Description

This Vertical Transverse Isotropy (VTI) Anisotropic Reverse Time Depth Migration (RTM) application measures the total time it takes to image 800 samples of various production size grids and write the final image to disk for the next work flow step involving 3-D seismic volume interpretation. In doing so, it reports the compute, interprocessor communication, and I/O performance of the individual functions that comprise the total solution. Unlike most references for the Reverse Time Migration, that focus solely on the performance of the 3D stencil compute kernel, this demonstration code additionally reports the total throughput involved in processing large data sets with a full 3D Anisotropic RTM application. It provides valuable insight into configuration and sizing for specific seismic processing requirements. The performance effects of new processors, interconnects, I/O subsystems, and software technologies can be evaluated while solving a real Exploration business problem.

This benchmark study uses the "in-core" implementation of this demonstration code where each node reads in only the trace, velocity, and conditioning data to be processed by that node plus a 4 element array pad (based on spatial order 8) shared with it's neighbors to the left and right during the initialization phase. It maintains previous, current, and next wavefield state information for each of the source, receiver, and anisotropic wavefields in memory. The second two grid dimensions used in this benchmark are specifically chosen to be prime numbers to exaggerate the effects of data alignment. Algorithm adaptions for processing higher orders in space and alternative "out-of-core" solutions using SSDs for wave state checkpointing are implemented in this demonstration application to better understand the effects of problem size scaling. Care is taken to handle absorption boundary conditioning and a variety of imaging conditions, appropriately.

RTM Application Structure:

Read Processing Parameter File, Determine Domain Decomposition, and Initialize Data Structures, and Allocate Memory.

Read Velocity, Epsilon, and Delta Data Based on Domain Decomposition and create source, receiver, & anisotropic previous, current, and next wave states.

First Loop over Time Steps

Compute 3D Stencil for Source Wavefield (a,s) - 8th order in space, 2nd order in time
Propagate over Time to Create s(t,z,y,x) & a(t,z,y,x)
Inject Estimated Source Wavelet
Apply Absorption Boundary Conditioning (a)
Update Wavefield States and Pointers
Write Snapshot of Wavefield (out-of-core) or Push Wavefield onto Stack (in-core)
Communicate Boundary Information

Second Loop over Time Steps
Compute 3D Stencil for Receiver Wavefield (a,r) - 8th order in space, 2nd order in time
Propagate over Time to Create r(t,z,y,x) & a(t,z,y,x)
Read Receiver Trace and Inject Receiver Wavelet
Apply Absorption Boundary Conditioning (a)
Update Wavefield States and Pointers
Communicate Boundary Information
Read in Source Wavefield Snapshot (out-of-core) or Pop Off of Stack (in-core)
Cross-correlate Source and Receiver Wavefields
Update image using image conditioning parameters

Write 3D Depth Image i(z,x,y) = Sum over time steps s(t,z,x,y) \* r(t,z,x,y) or other imaging conditions.

Key Points and Best Practices

This demonstration application represents a full Reverse Time Migration solution. Many references to the RTM application tend to focus on the compute kernel and ignore the complexity that the input, communication, and output bring to the task.

Image File MPI Write Performance Tuning

Changing the Image File Write from MPI non-blocking to MPI blocking and setting Oracle Message Passing Toolkit MPI environment variables revealed an 18x improvement in write performance to the Sun Storage 7410 system going from:

    86.8 to 4.8 seconds for the 1243 x 1151 x 1231 grid size
    183.1 to 10.2 seconds for the 2486 x 1151 x 1231 grid size

The Swat Sun Storage 7410 analytics data capture indicated an initial write performance of about 100 MB/sec with the MPI non-blocking implementation. After modifying to MPI blocking writes, Swat showed between 1.3 and 1.8 GB/sec with up to 13000 write ops/sec to write the final output image. The Swat results are consistent with the actual measured performance and provide valuable insight into the Reverse Time Migration application I/O performance.

The reason for this vast improvement has to do with whether the MPI file mode is sequential or not (MPI_MODE_SEQUENTIAL, O_SYNC, O_DSYNC). The MPI non-blocking routines, MPI_File_iwrite_at and MPI_wait, typically used for overlapping I/O and computation, do not support sequential file access mode. Therefore, the application could not take full performance advantages of the Sun Storage 7410 system write optimized cache. In contrast, the MPI blocking routine, MPI_File_write_at, defaults to MPI sequential mode and the performance advantages of the write optimized cache are realized. Since writing the final image is at the end of RTM execution, there is no need to overlap the I/O with computation.

Additional MPI parameters used:

    setenv SUNW_MP_PROCBIND true
    setenv MPI_SPIN 1
    setenv MPI_PROC_BIND 1

Adjusting the Level of Multithreading for Performance

The level of multithreading (8, 10, 12, 22, or 24) for various components of the RTM should be adjustable based on the type of computation taking place. Best to use OpenMP num_threads clause to adjust the level of multi-threading for each particular work task. Use numactl to specify how the threads are allocated to cores in accordance to the OpenMP parallelism level.

See Also

Disclosure Statement

Copyright 2010, Oracle and/or its affiliates. All rights reserved. Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names may be trademarks of their respective owners. Results as of 12/07/2010.

Sun Blade X6275 M2 Delivers Best Fluent (MCAE Application) Performance on Tested Configurations

This Manufacturing Engineering benchmark highlights the performance advantage the Sun Blade X6275 M2 server module offers over IBM, Cray, and SGI solutions as shown by the ANSYS FLUENT fluid dynamics application.

A cluster of eight of Oracle's Sun Blade X6275 M2 server modules delivered outstanding performance running the FLUENT 12 benchmark test suite.

  • The Sun Blade X6275 M2 server module cluster delivered the best results in all 36 of the test configurations run, outperforming the best posted results by as much as 42%.
  • The Sun Blade X6275 M2 server module demonstrated up to 76% performance improvement over the previous generation Sun Blade X6275 server module.

Performance Landscape

In the following tables, results are "Ratings" (bigger is better).
Rating = No. of sequential runs of test case possible in 1 day: 86,400/(Total Elapsed Run Time in Seconds)

The following table compares results on the basis of core count, irrespective of processor generation. This means that in some cases, i.e., for the 32-core and 64-core configurations, systems with the Intel Xeon X5670 six-core processors did not utilize quite all of the cores available for the specified processor count.


FLUENT 12 Benchmark Test Suite

Competitive Comparisons

System
Processors Cores Benchmark Test Case Ratings
eddy
417k
turbo
500k
aircraft
2m
sedan
4m
truck
14m
truck_poly
14m

Sun Blade X6275 M2 16 96 9340.5 39272.7 8307.7 8533.3 903.8 786.9
Best Posted 24 96

7562.4
797.0 712.9
Best Posted 16 96 7337.6 33553.4 6533.1 5989.6 739.1 683.5

Sun Blade X6275 M2 11 64 6306.6 27212.6 5592.2 5158.2 568.8 518.9
Best Posted 16 64 5556.3 26381.7 5494.4 4902.1 566.6 518.6

Sun Blade X6275 M2 8 48 4620.3 19093.9 4080.3 3251.2 376.0 359.4
Best Posted 8 48 4494.1 18989.0 3990.8 3185.3 372.7 354.5

Sun Blade X6275 M2 6 32 4061.1 15091.7 3275.8 3013.1 299.5 267.8
Best Posted 8 32 3404.9 14832.6 3211.9 2630.1 286.7 266.7

Sun Blade X6275 M2 4 24 2751.6 10441.1 2161.4 1907.3 188.2 182.5
Best Posted 6 24 1458.2 9626.7 1820.9 1747.2 185.1 180.8
Best Posted 4 24 2565.7 10164.7 2109.9 1608.2 187.1 180.8

Sun Blade X6275 M2 2 12 1429.9 5358.1 1097.5 813.2 95.9 95.9
Best Posted 2 12 1338.0 5308.8 1073.3 808.6 92.9 94.4



The following table compares results on the basis of processor count showing inter-generational processor performance improvement.


FLUENT 12 Benchmark Test Suite

Intergenerational Comparisons

System
Processors Cores Benchmark Test Case Ratings
eddy
417k
turbo
500k
aircraft
2m
sedan
4m
truck
14m
truck_poly
14m

Sun Blade X6275 M2 16 96 9340.5 39272.7 8307.7 8533.3 903.8 786.9
Sun Blade X6275 16 64 5308.8 26790.7 5574.2 5074.9 547.2 525.2
X6275 M2 : X6275 16
1.76 1.47 1.49 1.68 1.65 1.50

Sun Blade X6275 M2 8 48 4620.3 19093.9 4080.3 3251.2 376.0 359.4
Sun Blade X6275 8 32 3066.5 13768.9 3066.5 2602.4 289.0 270.3
X6275 M2 : X6275 8
1.51 1.39 1.33 1.25 1.30 1.33

Sun Blade X6275 M2 4 24 2751.6 10441.1 2161.4 1907.3 188.2 182.5
Sun Blade X6275 4 16 1714.3 7545.9 1519.1 1345.8 144.4 141.8
X6275 M2 : X6275 4
1.61 1.38 1.42 1.42 1.30 1.29

Sun Blade X6275 M2 2 12 1429.9 5358.1 1097.5 813.2 95.9 95.9
Sun Blade X6275 2 8 931.8 4061.1 827.2 681.5 73.0 73.8
X6275 M2 : X6275 2
1.53 1.32 1.33 1.19 1.31 1.30

Configuration Summary

Hardware Configuration:

8 x Sun Blade X6275 M2 server modules, each with
4 Intel Xeon X5670 2.93 GHz processors, turbo enabled
96 GB memory 1333 MHz
2 x 24 GB SATA-based Sun Flash Modules
2 x QDR InfiniBand Host Channel Adapter
Sun Datacenter InfiniBand Switch IB-36

Software Configuration:

Oracle Enterprise Linux Enterprise Server 5.5
ANSYS FLUENT V12.1.2
ANSYS FLUENT Benchmark Test Suite

Benchmark Description

The following description is from the ANSYS FLUENT website:

The FLUENT benchmarks suite comprises of a set of test cases covering a large range of mesh sizes, physical models and solvers representing typical industry usage. The cases range in size from a few 100 thousand cells to more than 100 million cells. Both the segregated and coupled implicit solvers are included, as well as hexahedral, mixed and polyhedral cell cases. This broad coverage is expected to demonstrate the breadth of FLUENT performance on a variety of hardware platforms and test cases.

The performance of a CFD code will depend on several factors, including size and topology of the mesh, physical models, numerics and parallelization, compilers and optimization, in addition to performance characteristics of the hardware where the simulation is performed. The principal objective of this benchmark suite is to provide comprehensive and fair comparative information of the performance of FLUENT on available hardware platforms.

About the ANSYS FLUENT 12 Benchmark Test Suite

    CFD models tend to be very large where grid refinement is required to capture with accuracy conditions in the boundary layer region adjacent to the body over which flow is occurring. Fine grids are required to also determine accurate turbulence conditions. As such these models can run for many hours or even days as well using a large number of processors.

Key Points and Best Practices

  • ANSYS FLUENT has not yet been certified by the vendor on Oracle Enterprise Linux (OEL). However, the ANSYS FLUENT benchmark tests have been run successfully on Oracle hardware running OEL as is (i.e. with NO changes or modifications).
  • The performance improvement of the Sun Blade X6275 M2 server module over the previous generation Sun Blade X6275 server module was due to two main factors: the increased core count per processor (6 vs. 4), and the more optimal, iterative dataset partitioning scheme used for the Sun Blade X6275 M2 server module.

See Also

Disclosure Statement

All information on the FLUENT website (http://www.fluent.com) is Copyrighted 1995-2010 by ANSYS Inc. Results as of December 06, 2010.

Tuesday Dec 07, 2010

Sun Blade X6275 M2 Server Module with Intel X5670 Processors SPEC CPU2006 Results

Results are presented for Oracle's Sun Blade X6275 M2 server module running the SPEC CPU2006 benchmark suite.
  • The dual-node Sun Blade X6275 M2 server module, equipped with two Intel Xeon X5670 2.93 GHz processors per node and running the Oracle Enterprise Linux 5.5 operating system delivered the best SPECint_rate2006 and SPECfp_rate2006 benchmark results for all systems with Intel Xeon processor 5000 sequence.

  • With a SPECint_rate2006 benchmark result of 679, the Sun Blade X6275 M2 server module, with two compute nodes per blade, delivers maximum performance for space constrained environments.

  • Comparing Oracle's dual-node blade to HP's dual-node blade server, based on their single node performance, the Sun Blade X6275 M2 server module SPECfp_rate2006 score of 241 outperforms the best published HP ProLiant BL2X220c G5 server score by 3.2x.

  • A single node of a Sun Blade X6275 M2 server module using 2.93 GHz Intel Xeon X5670 processors delivered 37% improvement in SPECint_rate2006 benchmark results and 22% improvement in SPECfp_rate2006 benchmark results compared to the previous generation Sun Blade X6275 server module.

  • Both nodes of a Sun Blade X6275 M2 server module using 2.93 GHz Intel Xeon X5670 processors delivered 59% improvement on the SPECint_rate2006 benchmark and 40% improvement on the SPECfp_rate2006 benchmark compared to the previous generation Sun Blade X6275 server module.

Performance Landscape

SPEC CPU2006 results comparing blade systems using Intel Xeon processor 5000 sequence based CPUs.

System SPECint_rate2006 SPECfp_rate2006
base peak base peak
Sun Blade X6275 M2 2-nodes, X5670 651 679 465 474
Sun Blade X6275 M2 1-node, X5670 326 348 234 241
IBM BladeCenter HS22V, X5680 352 377 246 254
Dell PowerEdge M710, X5680 355 380 247 256
HP BL460c G7, X5670 324 347 233 241
HP BL2X220c G5 1-node, E5450 106 132 67.4 74.8

SPEC CPU2006 results generational comparison between the Sun Blade X6275 M2 server module and the Sun Blade X6275 server module.

System SPECint_rate2006 SPECfp_rate2006
base peak base peak
Sun Blade X6275 M2 2-nodes, X5670 651 679 465 474
Sun Blade X6275 2-nodes, X5570 410 478 332 355
Sun Blade X6275 M2 1-node, X5670 326 348 234 241
Sun Blade X6275 1-node, X5570 238 253 191 197

Results in the above tables are from www.spec.org and this report as of 12/6/2010.

Configuration Summary and Results

Hardware Configuration:

Sun Blade X6275 server module, 2 nodes and each node has
2 x 2.93 GHz Intel Xeon X5670 processors, turbo enabled
96 GB, (12 x 8 GB DDR3-1333 DIMM)
Sun Storage 7410 System via NFS

Software Configuration:

Oracle Enterprise Linux Server release 5.5, kernel 2.6.18-194.el5
Intel 11.1
MicroQuill SmartHeap Library V8.1
SPEC CPU2006 V1.1

Results Summary:

Sun Blade X6275 M2, both nodes 651 SPECint_rate_base2006 679 SPECint_rate2006
Sun Blade X6275 M2, both nodes 465 SPECfp_rate_base2006 474 SPECfp_rate2006
Sun Blade X6275 M2, one node 326 SPECint_rate_base2006 348 SPECint_rate2006
Sun Blade X6275 M2, one node 234 SPECfp_rate_base2006 241 SPECfp_rate2006
Sun Blade X6275 M2, one node 36.2 SPECint_base2006 39.0 SPECint2006

Benchmark Description

SPEC CPU2006 is SPEC's most popular benchmark, with over 14000 results published in the years since it was introduced. It measures:

  • "Speed" - single copy performance of chip, memory, compiler
  • "Rate" - multiple copy (throughput)

The rate metrics are used for the throughput-oriented systems described on this page. These metrics include:

  • SPECint_rate2006: throughput for 12 integer benchmarks derived from real applications such as perl, gcc, XML processing, and pathfinding
  • SPECfp_rate2006: throughput for 17 floating point benchmarks derived from real applications, including chemistry, physics, genetics, and weather.

There are base variants of both the above metrics that require more conservative compilation. In particular, all benchmarks of a particular programming language must use the same compilation flags.

See Also

Disclosure Statement

SPEC and the benchmark names SPECint and SPECfp are registered trademarks of the Standard Performance Evaluation Corporation. Results are from the report and www.spec.org as of December 6, 2010.

Tuesday Oct 26, 2010

3D VTI Reverse Time Migration Scalability On Sun Fire X2270-M2 Cluster with Sun Storage 7210

This Oil & Gas benchmark shows the Sun Storage 7210 system delivers almost 2 GB/sec bandwidth and realizes near-linear scaling performance on a cluster of 16 Sun Fire X2270 M2 servers.

Oracle's Sun Storage 7210 system attached via QDR InfiniBand to a cluster of sixteen of Oracle's Sun Fire X2270 M2 servers was used to demonstrate the performance of a Reverse Time Migration application, an important application in the Oil & Gas industry. The total application throughput and computational kernel scaling are presented for two production sized grids of 800 samples.

  • Both the Reverse Time Migration I/O and combined computation shows near-linear scaling from 8 to 16 nodes on the Sun Storage 7210 system connected via QDR InfiniBand to a Sun Fire X2270 M2 server cluster:

      1243 x 1151 x 1231: 2.0x improvement
      2486 x 1151 x 1231: 1.7x improvement
  • The computational kernel of the Reverse Time Migration has linear to super-linear scaling from 8 to 16 nodes in Oracle's Sun Fire X2270 M2 server cluster:

      1243 x 1151 x 1231 : 2.2x improvement
      2486 x 1151 x 1231 : 2.0x improvement
  • Intel Hyper-Threading provides additional performance benefits to both the Reverse Time Migration I/O and computation when going from 12 to 24 OpenMP threads on the Sun Fire X2270 M2 server cluster:

      1243 x 1151 x 1231: 8% - computational kernel; 2% - total application throughput
      2486 x 1151 x 1231: 12% - computational kernel; 6% - total application throughput
  • The Sun Storage 7210 system delivers the Velocity, Epsilon, and Delta data to the Reverse Time Migration at a steady rate even when timing includes memory initialization and data object creation:

      1243 x 1151 x 1231: 1.4 to 1.6 GBytes/sec
      2486 x 1151 x 1231: 1.2 to 1.3 GBytes/sec

    One can see that when doubling the size of the problem, the additional complexity of overlapping I/O and multiple node file contention only produces a small reduction in read performance.

Performance Landscape

Application Scaling

Performance and scaling results of the total application, including I/O, for the reverse time migration demonstration application are presented. Results were obtained using a Sun Fire X2270 M2 server cluster with a Sun Storage 7210 system for the file server. The servers were running with hyperthreading enabled, allowing for 24 OpenMP threads per server.

Application Scaling Across Multiple Nodes
Number Nodes Grid Size - 1243 x 1151 x 1231 Grid Size - 2486 x 1151 x1231
Total Time (sec) Kernel Time (sec) Total Speedup Kernel Speedup Total Time (sec) Kernel Time (sec) Total Speedup Kernel Speedup
16 504 259 2.0 2.2\* 1024 551 1.7 2.0
14 565 279 1.8 2.0 1191 677 1.5 1.6
12 662 343 1.6 1.6 1426 817 1.2 1.4
10 784 394 1.3 1.4 1501 856 1.2 1.3
8 1024 560 1.0 1.0 1745 1108 1.0 1.0

\* Super-linear scaling due to the compute kernel fitting better into available cache

Application Scaling – Hyper-Threading Study

The affects of hyperthreading are presented when running the reverse time migration demonstration application. Results were obtained using a Sun Fire X2270 M2 server cluster with a Sun Storage 7210 system for the file server.

Hyper-Threading Comparison – 12 versus 24 OpenMP Threads
Number Nodes Thread per Node Grid Size - 1243 x 1151 x 1231 Grid Size - 2486 x 1151 x1231
Total Time (sec) Kernel Time (sec) Total HT Speedup Kernel HT Speedup Total Time (sec) Kernel Time (sec) Total HT Speedup Kernel HT Speedup
16 24 504 259 1.02 1.08 1024 551 1.06 1.12
16 12 515 279 1.00 1.00 1088 616 1.00 1.00

Read Performance

Read performance is presented for the velocity, epsilon and delta files running the reverse time migration demonstration application. Results were obtained using a Sun Fire X2270 M2 server cluster with a Sun Storage 7210 system for the file server. The servers were running with hyperthreading enabled, allowing for 24 OpenMP threads per server.

Velocity, Epsilon, and Delta File Read and Memory Initialization Performance
Number Nodes Overlap MBytes Read Grid Size - 1243 x 1151 x 1231 Grid Size - 2486 x 1151 x1231
Time (sec) Time Relative 8-node Total GBytes Read Read Rate GB/s Time (sec) Time Relative 8-node Total GBytes Read Read Rate GB/s
16 2040 16.7 1.1 23.2 1.4 36.8 1.1 44.3 1.2
8 951
14.8 1.0 22.1 1.6 33.0 1.0 43.2 1.3

Configuration Summary

Hardware Configuration:

16 x Sun Fire X2270 M2 servers, each with
2 x 2.93 GHz Intel Xeon X5670 processors
48 GB memory (12 x 4 GB at 1333 MHz)

Sun Storage 7210 system connected via QDR InfiniBand
2 x 18 GB SATA SSD (logzilla)
40 x 1 TB 7200 RM SATA disk

Software Configuration:

SUSE Linux Enterprise Server SLES 10 SP 2
Oracle Message Passing Toolkit 8.2.1 (for MPI)
Sun Studio 12 Update 1 C++, Fortran, OpenMP

Benchmark Description

This Reverse Time Migration (RTM) demonstration application measures the total time it takes to image 800 samples of various production size grids and write the final image to disk. In this version, each node reads in only the trace, velocity, and conditioning data to be processed by that node plus a four element inline 3-D array pad (spatial order of eight) shared with its neighbors to the left and right during the initialization phase. It represents a full RTM application including the data input, computation, communication, and final output image to be used by the next work flow step involving 3D volumetric seismic interpretation.

Key Points and Best Practices

This demonstration application represents a full Reverse Time Migration solution. Many references to the RTM application tend to focus on the compute kernel and ignore the complexity that the input, communication, and output bring to the task.

I/O Characterization without Optimal Checkpointing

Velocity, Epsilon, and Delta Files - Grid Reading

The additional amount of overlapping reads to share velocity, epsilon, and delta edge data with neighbors can be calculated using the following equation:

    (number_nodes - 1) x (order_in_space) x (y_dimension) x (z_dimension) x (4 bytes) x (3 files)

For this particular benchmark study, the additional 3-D pad overlap for the 16 and 8 node cases is:

    16 nodes: 15 x 8 x 1151 x 1231 x 4 x 3 = 2.04 GB extra
    8 nodes: 7 x 8 x 1151 x 1231 x 4 x 3 = 0.95 GB extra

For the first of the two test cases, the total size of the three files used for the 1243 x 1151 x 1231 case is

    1243 x 1151 x 1231 x 4 bytes = 7.05 GB per file x 3 files = 21.13 GB

With the additional 3-D pad, the total amount of data read is:

    16 nodes: 2.04 GB + 21.13 GB = 23.2 GB
    8 nodes: 0.95 GB + 21.13 GB = 22.1 GB

For the second of the two test cases, the total size of the three files used for the 2486 x 1151 x 1231 case is

    2486 x 1151 x 1231 x 4 bytes = 14.09 GB per file x 3 files = 42.27 GB

With the additional pad based on the number of nodes, the total amount of data read is:

    16 nodes: 2.04 GB + 42.27 GB = 44.3 GB
    8 nodes: 0.95 GB + 42.27 GB = 43.2 GB

Note that the amount of overlapping data read increases, not only by the number of nodes, but as the y dimension and/or the z dimension increases.

Trace Reading

The additional amount of overlapping reads to share trace edge data with neighbors for can be calculated using the following equation:

    (number_nodes - 1) x (order_in_space) x (y_dimension) x (4 bytes) x (number_of_time_slices)

For this particular benchmark study, the additional overlap for the 16 and 8 node cases is:

    16 nodes: 15 x 8 x 1151 x 4 x 800 = 442MB extra
    8 nodes: 7 x 8 x 1151 x 4 x 800 = 206MB extra

For the first case the size of the trace data file used for the 1243 x 1151 x 1231 case is

    1243 x 1151 x 4 bytes x 800 = 4.578 GB

With the additional pad based on the number of nodes, the total amount of data read is:

    16 nodes: .442 GB + 4.578 GB = 5.0 GB
    8 nodes: .206 GB + 4.578 GB = 4.8 GB

For the second case the size of the trace data file used for the 2486 x 1151 x 1231 case is

    2486 x 1151 x 4 bytes x 800 = 9.156 GB

With the additional pad based on the number of nodes, the total amount of data read is:

    16 nodes: .442 GB + 9.156 GB = 9.6 GB
    8 nodes: .206 GB + 9.156 GB = 9.4 GB

As the number of nodes is increased, the overlap causes more disk lock contention.

Writing Final Output Image

1243x1151x1231 - 7.1 GB per file:

    16 nodes: 78 x 1151 x 1231 x 4 = 442MB/node (7.1 GB total)
    8 nodes: 156 x 1151 x 1231 x 4 = 884MB/node (7.1 GB total)

2486x1151x1231 - 14.1 GB per file:

    16 nodes: 156 x 1151 x 1231 x 4 = 930 MB/node (14.1 GB total)
    8 nodes: 311 x 1151 x 1231 x 4 = 1808 MB/node (14.1 GB total)

Resource Allocation

It is best to allocate one node as the Oracle Grid Engine resource scheduler and MPI master host. This is especially true when running with 24 OpenMP threads in hyperthreading mode to avoid oversubscribing a node that is cooperating in delivering the solution.

See Also

Disclosure Statement

Copyright 2010, Oracle and/or its affiliates. All rights reserved. Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names may be trademarks of their respective owners. Results as of 10/20/2010.

Thursday Sep 30, 2010

Consolidation of 30 x86 Servers onto One SPARC T3-2

One of Oracle's SPARC T3-2 servers was able to consolidate the database workloads off of thirty older x86 servers in a secure virtualized environment.

  • The thirty x86 servers required 6.7 times more power than the consolidated workload on the SPARC T3-2 server.

  • The x86 configuration used 10 times the rack space than the consolidated workload did on the SPARC T3-2 server.

  • In addition to power & space considerations, there are also administrative cost savings resulting from having to manage just one server, as opposed to thirty servers.

  • Gartner says, "They need to realize that removing a single x86 server from a data center will result in savings of more than $400 a year in energy costs alone".

  • The total transaction throughput for the SPARC T3 server (132,000) was almost the same as the aggregate throughput achieved by the thirty x86 servers (138,000), where each x86 running at 10% utilization.

  • The average transaction response time on the SPARC T3-2 server (24 ms) was just a little higher than the average transaction response time on the Intel servers (19.5 ms).

Performance Landscape

System Oracle
Instances
Average
System
Utilization
Transactions/
min/system
Average
Response
time (ms)
watts/
system
OS
Sun Fire X4250
2x 3.0GHz Xeon
1 10% 4,600 19.5 320 Linux
SPARC T3-2
1x 1.65GHz SPARC T3
30 80% 132,000 24.0 1400\* Solaris

\* power consumption includes storage and periperal devices

Notes:
total throughput for 30 Intel systems = 30 \* 4600 = 138,000
total watts for 30 Intel systems = 30 \* 320 = 9600

Results and Configuration Summary

x86 Server Configuration:

30 x Sun Fire X4250 servers, each with
2 X Intel 3.0 GHz E5450 processors
16 GB memory
6 x internal 146 GB 15K SAS disks
RedHat Linux 5.3
Oracle Database 11g Release 2

SPARC T3 Server Configuration:

1 x SPARC T3-2 server
2 x 1.65 GHz SPARC T3 processors
256 GB memory
2 X 10K 300 GB internal SAS disks
1 x Sun Storage F5100 Flash Array storage
1 x Sun Fires X4270 server as COMSTAR target
Oracle Solaris 10 9/10
Oracle Database 11g Release 2

Benchmark Description

This demonstration was designed to show the benefits of virtualization when upgrading from older X86 systems to one of Oracle's T-series servers. A 30:1 consolidation was shown moving from thirty X86 Linux servers to a single T-Series server running Oracle Solaris in a secure virtualized environment. After the consolidation, there was still 20% headroom in the SPARC T3-2 server for additional growth in the workload.

The 200 scale iGen OLTP workload was used to test the consolidation. The x86 system was loaded with iGen clients up to a level of 10% cpu utilization. This load level for x86 systems is typically found in many data centers.

Thirty Oracle Solaris zones (containers) were created on the SPARC T3-2 server, with each zone configured identically as the Oracle configuration on the x86 server. The throughput on each zone was ramped up to the same level as on the Intel base server.

The overall CPU utilization on the SPARC T3-2 server, together with the average iGen transaction response times were then measured along with the power consumption.

Key Points and Best Practices

  • Each Oracle Solaris container was assigned to a processor set consisting of eight virtual CPUs. This use of processor sets was critical to obtaining the reported performance number. Without processor set, the performance was reduced to about one-half the reported performance number.

  • Once the first container was completely configured (with Oracle 11g and iGen installed), the remaining containers were created by a simple cloning procedure, which took no more than a few minutes for each container.

  • Setting up a standalone x86 server with Linux, Oracle and iGen is a far more time consuming task than setting up additional containers once the first container has been created.

See Also

Disclosure Statement

Copyright 2010, Oracle and/or its affiliates. All rights reserved. Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names may be trademarks of their respective owners. Results as of 9/20/2010.

Monday Sep 27, 2010

Sun Fire X2270 M2 Super-Linear Scaling of Hadoop Terasort and CloudBurst Benchmarks

A 16-node cluster of Oracle's Sun Fire X2270 M2 servers showed super-linear scaling of two Hadoop benchmarks. Performance was measured using the Terasort benchmark with a 100GB data set. In addition, performance was measured using Cloudburst which maps next generation "short read" sequence data onto the human and other genomes.

  • On the Terasort workload, a 16-node Sun Fire X2270 M2 cluster sorted the 100GB data set at a rate of 433.3 MB/s finishing in 236.3 seconds.

  • The 16-node Sun Fire X2270 M2 cluster was 9.3x faster on a per node basis than the 2010 winner of the Terasort benchmark competition (www.sortbenchmark.org) which used a 3,452-node Xeon cluster to sort 100 TB of input data in 173 minutes. Both systems used Hadoop, Terasort and 2-socket x86 servers. Allowances have to be made for the differences in problem complexity.

  • The Terasort benchmark showed super-linear scaling on the Sun Fire X2270 M2 cluster (total of 32 Intel 2.93GHz Xeons).

  • Using Cloudburst on a workload of the human genome and the SRR001113 short read data set, a 16-node Sun Fire X2270 M2 cluster finished mapping the short reads onto the human genome in 34.2 minutes.

  • On a per node basis, a 2-node Sun Fire X2270 M2 cluster was 1.7x faster than a 12-node Xeon cluster that processed the human genome and the SRR001113 short read data set in approximately 60,000 seconds (see figure 3 of this journal article). Both systems used Hadoop, CloudBurst and x86 servers.

  • The Terasort benchmark showed super-linear scaling on the Sun Fire X2270 M2 cluster (total of 32 Intel 2.93GHz Xeons).

Performance Landscape

Terasort
100 GB input data set
Performance is "real" execution time reported by /usr/bin/time in seconds (smaller is better)
Number
of Nodes
Seconds Scaling Linear
Scaling
16 236.3 25.4 16
8 466.3 12.9 8
4 927.2 6.5 4
2 2140.8 2.8 2
1 6010.2 1.0 1

CloudBurst
SR001111 short read data set mapped onto the hs_ref_GRCh37 human genome
Performance is "total running" time reported by CloudBurst in seconds (smaller is better)
Number
of Nodes
Seconds Scaling Linear
Scaling
16 2054.9 8.4 8
8 3615.8 4.7 4
4 7895.7 2.2 2
2 17155.1 1.0 1

Results and Configuration Summary

Hardware Configuration:

16 x Sun Fire X2270 M2 server, each server with
2 Intel Xeon X5670 2.93GHz processors, turbo enabled
96 GB memory 1066 MHz
HDD SATA 1 TB 7200 RPM 3.5-in.
2 x 10/100/1000 ethernet

Software Configuration:

Oracle Solaris 10 10/09
Java Platform, Standard Edition, JDK 6 Update 20 Performance Release
Hadoop v0.20.2

Benchmark Description

The Apache Hadoop middleware is the Yahoo implementation of Google's Map Reduce. Map Reduce permits the programmer to write serial code that Map Reduce schedules for parallel execution. Map Reduce has been applied to a wide variety of problems, including image processing, sorting, database merging and genomics.

Hadoop uses the Hadoop Distributed Filesystem (HDFS) that distributes data across the local disks of a cluster such that each node in the cluster accesses its local disk to the greatest extent possible.

Results for two different Hadoop benchmarks are reported above:

  • Terasort is an I/O intensive benchmark that was originally developed by Jim Gray. By having many Hadoop data nodes, it is possible to achieve high I/O capacity. For purposes of benchmarking, the Teragen program was used to create an input data set that comprised 100 GB.

  • CloudBurst is a genome assembly benchmark that was developed by Michael Schatz, previously of the University of Maryland and presently of Cold Springs Harbor Laboratory. CloudBurst maps what is known as DNA short read data onto a reference genome. For purposes of benchmarking, the SRR001113 short read data set is mapped onto the hs_ref_GRCh37 sequence data for all chromosomes of the human genome. Specifically, the hs_ref_GRCh37 FASTA files for chromosomes 1, 2, ... 21, 22, X and Y were catenated in that order to obtain one large FASTA file that represented all chromosomes of the human genome. For purposes of benchmarking, any DNA fragment from the SRR001113 short read data set that contained more than three mismatches was ignored.

See Also

Disclosure Statement

Hadoop, see http://hadoop.apache.org/ for more information. Results as of 9/20/2010.

Tuesday Sep 21, 2010

ProMAX Performance and Throughput on Sun Fire X2270 and Sun Storage 7410

Halliburton/Landmark's ProMAX 3D Prestack Kirchhoff Time Migration's single job scalability and multiple job throughput using various scheduling methods are evaluated on a cluster of Oracle's Sun Fire X2270 servers attached via QDR InfiniBand to Oracle's Sun Storage 7410 system.

Two resource scheduling methods, compact and distributed, are compared while increasing the system load with additional concurrent ProMAX jobs.

  • A single ProMAX job has near linear scaling of 5.5x on 6 nodes of a Sun Fire X2270 cluster.

  • A single ProMAX job has near linear scaling of 7.5x on a Sun Fire X2270 server when running from 1 to 8 threads.

  • ProMAX can take advantage of Oracle's Sun Storage 7410 system features compared to dedicated local disks. There was no significant difference in run time observed when running up to 8 concurrent 16 thread jobs.

  • The 8-thread ProMAX job throughput using the distributed scheduling method is equivalent or slightly faster than the compact scheme for 1 to 4 concurrent jobs.

  • The 16-thread ProMAX job throughput using the distributed scheduling method is up to 8% faster when compared to the compact scheme on an 8-node Sun Fire X2270 cluster.

The multiple job throughput characterization revealed in this benchmark study are key in pre-configuring Oracle Grid Engine resource scheduling for ProMAX on a Sun Fire X2270 cluster and provide valuable insight for server consolidation.

Performance Landscape

Single Job Scaling

Single job performance on a single node is near linear up the number of cores in the node, i.e. 2 Intel Xeon X5570s with 4 cores each. With hyperthreading (2 active threads per core) enabled, more ProMAX threads are used increasing the load on the CPU's memory architecture causing the reduced speedups.
ProMAX single job performance on the 6-node cluster shows near linear speedup node to node.
Single Job 6-Node Scalability
Hyperthreading Enabled - 16 Threads/Node Maximum
Number of Nodes Threads Per Node Speedup to 1 Thread Speedup to 1 Node
6 16 54.2 5.5
4 16 36.2 3.6
3 16 26.1 2.6
2 16 17.6 1.8
1 16 10.0 1.0
1 14 9.2
1 12 8.6
1 10 7.2\*
1 8 7.5
1 6 5.9
1 4 3.9
1 3 3.0
1 2 2.0
1 1 1.0

\* 2 threads contend with two master node daemons

Multiple Job Throughput Scaling, Compact Scheduling

With the Sun Storage 7410 system, performance of 8 concurrent jobs on the cluster using compact scheduling is equivalent to a single job.

Multiple Job Throughput Scalability
Hyperthreading Enabled - 16 Threads/Node Maximum
Number of Nodes Number of Nodes per Job Threads Per Node per Job Performance Relative to 1 Job Total Nodes Percent Cluster Used
1 1 16 1.00 1 13
2 1 16 1.00 2 25
4 1 16 1.00 4 50
8 1 16 1.00 8 100

Multiple 8-Thread Job Throughput Scaling, Compact vs. Distributed Scheduling

These results report the difference of different distributed method resource scheduling levels to 1, 2, and 4 concurrent job compact method baselines.

Multiple 8-Thread Job Scheduling
HyperThreading Enabled - Use 8 Threads/Node Maximum
Number of Jobs Number of Nodes per Job Threads Per Node per Job Performance Relative to 1 Job Total Nodes Total Threads per Node Used Percent of PVM Master 8 Threads Used
1 1 8 1.00 1 8 100
1 4 2 1.01 4 2 25
1 8 1 1.01 8 1 13

2 1 8 1.00 2 8 100
2 4 2 1.01 4 4 50
2 8 1 1.01 8 2 25

4 1 8 1.00 4 8 100
4 4 2 1.00 4 8 100
4 8 1 1.01 8 4 100

Multiple 16-Thread Job Throughput Scaling, Compact vs. Distributed Scheduling

The results are reported relative to the performance of 1, 2, 4, and 8 concurrent 2-node, 8-thread jobs.

Multiple 16-Thread Job Scheduling
HyperThreading Enabled - 16 Threads/Node Available
Number of Jobs Number of Nodes per Job Threads Per Node per Job Performance Relative to 1 Job Total Nodes Total Threads per Node Used Percent of PVM Master 16 Threads Used
1 1 16 0.66 1 16 100\*
1 2 8 1.00 2 8 50
1 4 4 1.03 4 4 25
1 8 2 1.06 8 2 13

2 1 16 0.70 2 16 100\*
2 2 8 1.00 4 8 50
2 4 4 1.07 8 4 25
2 8 2 1.08 8 4 25

4 1 16 0.74 4 16 100\*
4 4 4 0.74 4 16 100\*
4 2 8 1.00 8 8 50
4 4 4 1.05 8 8 50
4 8 2 1.04 8 8 50

8 1 16 1.00 8 16 100\*
8 4 4 1.00 8 16 100\*
8 8 2 1.00 8 16 100\*

\* master PVM host; running 20 to 21 total threads (over-subscribed)

Results and Configuration Summary

Hardware Configuration:

8 x Sun Fire X2270 servers, each with
2 x 2.93 GHz Intel Xeon X5570 processors
48 GB memory at 1333 MHz
1 x 500 GB SATA
Sun Storage 7410 system
4 x 2.3 GHz AMD Opteron 8356 processors
128 GB memory
2 Internal 233GB SAS drives = 466 GB
2 Internal 93 GB read optimized SSD = 186 GB
1 External Sun Storage J4400 array with 22 1TB SATA drives and 2 18GB write optimized SSD
11 TB mirrored data and mirrored write optimized SSD

Software Configuration:

SUSE Linux Enterprise Server 10 SP 2
Parallel Virtual Machine 3.3.11
Oracle Grid Engine
Intel 11.1 Compilers
OpenWorks Database requires Oracle 10g Enterprise Edition
Libraries: pthreads 2.4, Java 1.6.0_01, BLAS, Stanford Exploration Project Libraries

Benchmark Description

The ProMAX family of seismic data processing tools is the most widely used Oil and Gas Industry seismic processing application. ProMAX is used for multiple applications, from field processing and quality control, to interpretive project-oriented reprocessing at oil companies and production processing at service companies. ProMAX is integrated with Halliburton's OpenWorks Geoscience Oracle Database to index prestack seismic data and populate the database with processed seismic.

This benchmark evaluates single job scalability and multiple job throughput of the ProMAX 3D Prestack Kirchhoff Time Migration while processing the Halliburton benchmark data set containing 70,808 traces with 8 msec sample interval and trace length of 4992 msec. Alternative thread scheduling methods are compared for optimizing single and multiple job throughput. The compact scheme schedules the threads of a single job in as few nodes as possible, whereas, the distributed scheme schedules the threads across a many nodes as possible. The effects of load on the Sun Storage 7410 system are measured. This information provides valuable insight into determining the Oracle Grid Engine resource management policies.

Hyperthreading is enabled for all of the tests. It should be noted that every node is running a PVM daemon and ProMAX license server daemon. On the master PVM daemon node, there are three additional ProMAX daemons running.

The first test measures single job scalability across a 6-node cluster with an additional node serving as the master PVM host. The speedup relative to a single node, single thread are reported.

The second test measures multiple job scalability running 1 to 8 concurrent 16-thread jobs using the Sun Storage 7410 system. The performance is reported relative to a single job.

The third test compares 8-thread multiple job throughput using different job scheduling methods on a cluster. The compact method involves putting all 8 threads for a job on the same node. The distributed method involves spreading the 8 threads of job across multiple nodes. The results report the difference of different distributed method resource scheduling levels to 1, 2, and 4 concurrent job compact method baselines.

The fourth test is similar to the second test except running 16-thread ProMAX jobs. The results are reported relative to the performance of 1, 2, 4, and 8 concurrent 2-node, 8-thread jobs.

The ProMAX processing parameters used for this benchmark:

Minimum output inline = 65
Maximum output inline = 85
Inline output sampling interval = 1
Minimum output xline = 1
Maximum output xline = 200 (fold)
Xline output sampling interval = 1
Antialias inline spacing = 15
Antialias xline spacing = 15
Stretch Mute Aperature Limit with Maximum Stretch = 15
Image Gather Type = Full Offset Image Traces
No Block Moveout
Number of Alias Bands = 10
3D Amplitude Phase Correction
No compression
Maximum Number of Cache Blocks = 500000

Key Points and Best Practices

  • The application was rebuilt with the Intel 11.1 Fortran and C++ compilers with these flags.

    -xSSE4.2 -O3 -ipo -no-prec-div -static -m64 -ftz -fast-transcendentals -fp-speculation=fast
  • There are additional execution threads associated with a ProMAX node. There are two threads that run on each node: the license server and PVM daemon. There are at least three additional daemon threads that run on the PVM master server: the ProMAX interface GUI, the ProMAX job execution - SuperExec, and the PVM console and control. It is best to allocate one node as the master PVM server to handle the additional 5+ threads. Otherwise, hyperthreading can be enabled and the master PVM host can support up to 8 ProMAX job threads.

  • When hyperthreading is enabled in on one of the non-master PVM hosts, there is a 7% penalty going from 8 to 10 threads. However, 12 threads are 11 percent faster than 8. This can be contributed to the two additional support threads when hyperthreading initiates.

  • Single job performance on a single node is near linear up the number of cores in the node, i.e. 2 Intel Xeon X5570s with 4 cores each. With hyperthreading (2 active threads per core) enabled, more ProMAX threads are used increasing the load on the CPU's memory architecture causing the reduced speedups.

    Users need to be aware of these performance differences and how it effects their production environment.

See Also

Disclosure Statement

The following are trademarks or registered trademarks of Halliburton/Landmark Graphics: ProMAX. Results as of 9/20/2010.

Monday Sep 20, 2010

Schlumberger's ECLIPSE 300 Performance Throughput On Sun Fire X2270 Cluster with Sun Storage 7410

Oracle's Sun Storage 7410 system, attached via QDR InfiniBand to a cluster of eight of Oracle's Sun Fire X2270 servers, was used to evaluate multiple job throughput of Schlumberger's Linux-64 ECLIPSE 300 compositional reservoir simulator processing their standard 2 Million Cell benchmark model with 8 rank parallelism (MM8 job).

  • The Sun Storage 7410 system showed little difference in performance (2%) compared to running the MM8 job with dedicated local disk.

  • When running 8 concurrent jobs on 8 different nodes all to the Sun Storage 7140 system, the performance saw little degradation (5%) compared to a single MM8 job running on dedicated local disk.

Experiments were run changing how the cluster was utilized in scheduling jobs. Rather than running with the default compact mode, tests were run distributing the single job among the various nodes. Performance improvements were measured when changing from the default compact scheduling scheme (1 job to 1 node) to a distributed scheduling scheme (1 job to multiple nodes).

  • When running at 75% of the cluster capacity, distributed scheduling outperformed the compact scheduling by up to 34%. Even when running at 100% of the cluster capacity, the distributed scheduling is still slightly faster than compact scheduling.

  • When combining workloads, using the distributed scheduling allowed two MM8 jobs to finish 19% faster than the reference time and a concurrent PSTM workload to find 2% faster.

The Oracle Solaris Studio Performance Analyzer and Sun Storage 7410 system analytics were used to identify a 3D Prestack Kirchhoff Time Migration (PSTM) as a potential candidate for consolidating with ECLIPSE. Both scheduling schemes are compared while running various job mixes of these two applications using the Sun Storage 7410 system for I/O.

These experiments showed a potential opportunity for consolidating applications using Oracle Grid Engine resource scheduling and Oracle Virtual Machine templates.

Performance Landscape

Results are presented below on a variety of experiments run using the 2009.2 ECLIPSE 300 2 Million Cell Performance Benchmark (MM8). The compute nodes are a cluster of Sun Fire X2270 servers connected with QDR InfiniBand. First, some definitions used in the tables below:

Local HDD: Each job runs on a single node to its dedicated direct attached storage.
NFSoIB: One node hosts its local disk for NFS mounting to other nodes over InfiniBand.
IB 7410: Sun Storage 7410 system over QDR InfiniBand.
Compact Scheduling: All 8 MM8 MPI processes run on a single node.
Distributed Scheduling: Allocate the 8 MM8 MPI processes across all available nodes.

First Test

The first test compares the performance of a single MM8 test on a single node using local storage to running a number of jobs across the cluster and showing the effect of different storage solutions.

Compact Scheduling
Multiple Job Throughput Results Relative to Single Job
2009.2 ECLIPSE 300 MM8 2 Million Cell Performance Benchmark

Cluster Load Number of MM8 Jobs Local HDD Relative Throughput NFSoIB Relative Throughput IB 7410 Relative Throughput
13% 1 1.00 1.00\* 0.98
25% 2 0.98 0.97 0.98
50% 4 0.98 0.96 0.97
75% 6 0.98 0.95 0.95
100% 8 0.98 0.95 0.95

\* Performance measured on node hosting its local disk to other nodes in the cluster.

Second Test

This next test uses the Sun Storage 7410 system and compares the performance of running the MM8 job on 1 node using the compact scheduling to running multiple jobs with compact scheduling and to running multiple jobs with the distributed schedule. The tests are run on a 8 node cluster, so each distributed job has only 1 MPI process per node.

Comparing Compact and Distributed Scheduling
Multiple Job Throughput Results Relative to Single Job
2009.2 ECLIPSE 300 MM8 2 Million Cell Performance Benchmark

Cluster Load Number of MM8 Jobs Compact Scheduling
Relative Throughput
Distributed Scheduling\*
Relative Throughput
13% 1 1.00 1.34
25% 2 1.00 1.32
50% 4 0.99 1.25
75% 6 0.97 1.10
100% 8 0.97 0.98

\* Each distributed job has 1 MPI process per node.

Third Test

This next test uses the Sun Storage 7410 system and compares the performance of running the MM8 job on 1 node using the compact scheduling to running multiple jobs with compact scheduling and to running multiple jobs with the distributed schedule. This test only uses 4 nodes, so each distributed job has two MPI processes per node.

Comparing Compact and Distributed Scheduling on 4 Nodes
Multiple Job Throughput Results Relative to Single Job
2009.2 ECLIPSE 300 MM8 2 Million Cell Performance Benchmark

Cluster Load Number of MM8 Jobs Compact Scheduling
Relative Throughput
Distributed Scheduling\*
Relative Throughput
25% 1 1.00 1.39
50% 2 1.00 1.28
100% 4 1.00 1.00

\* Each distributed job it has two MPI processes per node.

Fourth Test

The last test involves running two different applications on the 4 node cluster. It compares the performance of running the cluster fully loaded and changing how the applications are run, either compact or distributed. The comparisons are made against the individual application running the compact strategy (as few nodes as possible). It shows that appropriately mixing jobs can give better job performance than running just one kind of application on a single cluster.

Multiple Job, Multiple Application Throughput Results
Comparing Scheduling Strategies
2009.2 ECLIPSE 300 MM8 2 Million Cell and 3D Kirchoff Time Migration (PSTM)

Number of PSTM Jobs Number of MM8 Jobs Compact Scheduling
(1 node x 8 processes
per job)
ECLIPSE
Distributed Scheduling
(4 nodes x 2 processes
per job)
ECLIPSE
Distributed Scheduling
(4 nodes x 4 processes
per job)
PSTM
Compact Scheduling
(2 nodes x 8 processes per job)
PSTM
Cluster Load
0 1 1.00 1.40

25%
0 2 1.00 1.27

50%
0 4 0.99 0.98

100%
1 2
1.19 1.02
100%
2 0

1.07 0.96 100%
1 0

1.08 1.00 50%

Results and Configuration Summary

Hardware Configuration:

8 x Sun Fire X2270 servers, each with
2 x 2.93 GHz Intel Xeon X5570 processors
24 GB memory (6 x 4 GB memory at 1333 MHz)
1 x 500 GB SATA
Sun Storage 7410 system, 24 TB total, QDR InfiniBand
4 x 2.3 GHz AMD Opteron 8356 processors
128 GB memory
2 Internal 233GB SAS drives (466 GB total)
2 Internal 93 GB read optimized SSD (186 GB total)
1 Sun Storage J4400 with 22 1 TB SATA drives and 2 18 GB write optimized SSD
20 TB RAID-Z2 (double parity) data and 2-way striped write optimized SSD or
11 TB mirrored data and mirrored write optimized SSD
QDR InfiniBand Switch

Software Configuration:

SUSE Linux Enterprise Server 10 SP 2
Scali MPI Connect 5.6.6
GNU C 4.1.2 compiler
2009.2 ECLIPSE 300
ECLIPSE license daemon flexlm v11.3.0.0
3D Kirchoff Time Migration

Benchmark Description

The benchmark is a home-grown study in resource usage options when running the Schlumberger ECLIPSE 300 Compositional reservoir simulator with 8 rank parallelism (MM8) to process Schlumberger's standard 2 Million Cell benchmark model. Schlumberger pre-built executables were used to process a 260x327x73 (2 Million Cell) sub-grid with 6,206,460 total grid cells and model 7 different compositional components within a reservoir. No source code modifications or executable rebuilds were conducted.

The ECLIPSE 300 MM8 job uses 8 MPI processes. It can run within a single node (compact) or across multiple nodes of a cluster (distributed). By using the MM8 job, it is possible to compare the performance between running each job on a separate node using local disk to using a shared network attached storage solution. The benchmark tests study the affect of increasing the number of MM8 jobs in a throughput model.

The first test compares the performance of running 1, 2, 4, 6 and 8 jobs on a cluster of 8 nodes using local disk, NFSoIB disk, and the Sun Storage 7410 system connected via InfiniBand. Results are compared against the time it takes to run 1 job with local disk. This test shows what performance impact there is when loading down a cluster.

The second test compares different methods of scheduling jobs on a cluster. The compact method involves putting all 8 MPI processes for a job on the same node. The distributed method involves using 1 MPI processes per node. The results compare the performance against 1 job on one node.

The third test is similar to the second test, but uses only 4 nodes in the cluster, so when running distributed, there are 2 MPI processes per node.

The fourth test compares the compact and distributed scheduling methods on 4 nodes while running a 2 MM8 jobs and one 16-way parallel 3D Prestack Kirchhoff Time Migration (PSTM).

Key Points and Best Practices

  • ECLIPSE is very sensitive to memory bandwidth and needs to be run on 1333 MHz or greater memory speeds. In order to maintain 1333 MHz memory, the maximum memory configuration for the processors used in this benchmark is 24 GB. Bios upgrades now allow 1333 MHz memory for up to 48 GB of memory. Additional nodes can be used to handle data sets that require more memory than available per node. Allocating at least 20% of memory per node for I/O caching helps application performance.

  • If allocating an 8-way parallel job (MM8) to a single node, it is best to use an ECLIPSE license for that particular node to avoid the any additional network overhead of sharing a global license with all the nodes in a cluster.

  • Understanding the ECLIPSE MM8 I/O access patterns is essential to optimizing a shared storage solution. The analytics available on the Oracle Unified Storage 7410 provide valuable I/O characterization information even without source code access. A single MM8 job run shows an initial read and write load related to reading the input grid, parsing Petrel ascii input parameter files and creating an initial solution grid and runtime specifications. This is followed by a very long running simulation that writes data, restart files, and generates reports to the 7410. Due to the nature of the small block I/O, the mirrored configuration for the 7410 outperformed the RAID-Z2 configuration.

    A single MM8 job reads, processes, and writes approximately 240 MB of grid and property data in the first 36 seconds of execution. The actual read and write of the grid data, that is intermixed with this first stage of processing, is done at a rate of 240 MB/sec to the 7410 for each of the two operations.

    Then, it calculates and reports the well connections at an average 260 KB writes/second with 32 operations/second = 32 x 8 KB writes/second. However, the actual size of each I/O operation varies between 2 to 100 KB and there are peaks every 20 seconds. The write cache is on average operating at 8 accesses/second at approximately 61 KB/second (8 x 8 KB writes/sec). As the number of concurrent jobs increases, the interconnect traffic and random I/O operations per second to the 7410 increases.

  • MM8 multiple job startup time is reduced on shared file systems, if each job uses separate input files.

See Also

Disclosure Statement

Copyright 2010, Oracle and/or its affiliates. All rights reserved. Oracle and Java are registered trademarks of Oracle and/or its affiliates. Other names may be trademarks of their respective owners. Results as of 9/20/2010.

Sun Fire X4470 4 Node Cluster Delivers World Record SAP SD-Parallel Benchmark Result

Oracle delivered an SAP enhancement package 4 for SAP ERP 6.0 Sales and Distribution – Parallel (SD-Parallel) Benchmark world record result using four of Oracle's Sun Fire X4470 servers, Oracle Solaris 10 and Oracle 11g Real Application Clusters (RAC) software.

  • The Sun Fire X4470 servers delivered 8% more performance compared to the IBM Power 780 server running the SAP enhancement package 4 for SAP ERP 6.0 Sales and Distribution benchmark.

  • The Sun Fire X4470 servers result of 40,000 users delivered 2.2 times the performance of the HP ProLiant DL980 G7 result of 18,180 users.

  • The Sun Fire X4470 servers result of 40,000 users delivered 2.5 times the performance of the Fujitsu PRIMEQUEST 1800E result of 16,000 users.

This result shows that a complete software and hardware solution from Oracle using Oracle RAC, Oracle Solaris and Sun servers provides a superior performing solution.

Performance Landscape

Selected SAP Sales and Distribution benchmark results are presented in decreasing order in performance. All benchmarks were using SAP enhancement package 4 for SAP ERP 6.0 (Unicode) except the result marked with an asterix (\*) which was achieved with SAP ERP 6.0.

System OS
Database
Users SAPS Type Date
Four Sun Fire X4470
4xIntel Xeon X7560 @2.26GHz
256 GB
Solaris 10
Oracle 11g Real Application Clusters
40,000 221,014 Parallel 20-Sep-10
Five IBM System p 570 (\*)
8xPOWER6 @4.7GHz
128 GB
AIX 5L Version 5.3
Oracle 10g Real Application Clusters
37,040 187,450 Parallel "non-Unicode" 25-Mar-08
IBM Power 780
8xPOWER7 @3.8GHz
1 TB
AIX 6.1
DB2 9.7
37,000 202,180 2-Tier 7-Apr-10
Two Sun Fire X4470
4xIntel Xeon X7560 @2.26GHz
256 GB
Solaris 10
Oracle 11g Real Application Clusters
21,000 115,300 Parallel 28-Jun-10
HP DL980 G7
8xIntel Xeon X7560 @2.26GHz
512 GB
Win Server 2008 R2 DE
SQL Server 2008
18,180 99,320 2-Tier 21-Jun-10
Fujitsu PRIMEQUEST 1800E
8xIntel Xeon X7560 @2.26GHz
512 GB
Win Server 2008 R2 DE
SQL Server 2008
16,000 87,550 2-Tier 30-Mar-10
Four Sun Blade X6270
2xIntel Xeon X5570 @2.93GHz
48 GB
Solaris 10
Oracle 10g Real Application Clusters
13,718 75,762 Parallel 12-Oct-09
HP DL580 G7
4xIntel Xeon X7560 @2.26GHz
256 GB
Win Server 2008 R2 DE
SQL Server 2008
10,445 57,020 2-Tier 21-Jun-10
Two Sun Blade X6270
2xIntel Xeon X5570 @2.93GHz
48 GB
Solaris 10
Oracle 10g Real Application Clusters
7,220 39,420 Parallel 12-Oct-09
One Sun Blade X6270
2xIntel Xeon X5570 @2.93GHz
48 GB
Solaris 10
Oracle 10g Real Application Clusters
3,800 20,750 Parallel 12-Oct-09

Complete benchmark results and a description can be found at the SAP benchmark website http://www.sap.com/solutions/benchmark/sd.epx.

Results and Configuration Summary

Hardware Configuration:

4 x Sun Fire X4470 servers, each with
4 x Intel Xeon X7560 2.26 GHz (4 chips, 32 cores, 64 threads)
256 GB memory

Software Configuration:

Oracle 11g Real Application Clusters (RAC)
Oracle Solaris 10

Results Summary:

Number of SAP SD benchmark users:
40,000
Average dialog response time:
0.86 seconds
Throughput:

Dialog steps/hour:
13,261,000

SAPS:
221,020
SAP Certification:
2010039

Benchmark Description

SAP is one of the premier world-wide ERP application providers and maintains a suite of benchmark tests to demonstrate the performance of competitive systems running the various SAP products.

The SAP Standard Application SD Benchmark represents the critical tasks performed in real-world ERP business environments. The SAP Standard Application Sales and Distribution - Parallel (SD-Parallel) Benchmark is a two-tier ERP business test that is indicative of full business workloads of complete order processing and invoice processing and demonstrates the ability to run both the application and database software on a single system.

The SD-Parallel Benchmark consists of the same transactions and user interaction steps as the SD Benchmark. This means that the SD-Parallel Benchmark runs the same business processes as the SD Benchmark. The difference between the benchmarks is the technical data distribution.

The additional rule for parallel and distributed databases is one must equally distribute the benchmark users across all database nodes for the used benchmark clients (round-robin method). Following this rule, all database nodes work on data of all clients. This avoids unrealistic configurations such as having only one client per database node.

The SAP Benchmark Council agreed to give the parallel benchmark a different name so that the difference can be easily recognized by any interested parties - customers, prospects, and analysts. The naming convention is SD-Parallel for Sales & Distribution - Parallel.

In January 2009, a new version of the SAP enhancement package 4 for SAP ERP 6.0 (Unicode) Sales and Distribution (SD) Benchmark was released. This new release has higher cpu requirements and so yields from 25-50% fewer users compared to the previous (non-unicode) Standard Sales and Distribution (SD) Benchmark. Between 10-30% of this greater load is due to the extra overhead from the processing of the larger character strings due to Unicode encoding.

Unicode is a computing standard that allows for the representation and manipulation of text expressed in most of the world's writing systems. Before the Unicode requirement, this benchmark used ASCII characters meaning each was just 1 byte. The new version of the benchmark requires Unicode characters and the Application layer (where ~90% of the cycles in this benchmark are spent) uses a new encoding, UTF-16, which uses 2 bytes to encode most characters (including all ASCII characters) and 4 bytes for some others. This requires computers to do more computation and use more bandwidth and storage for most character strings. Refer to the above SAP Note for more details.

See Also

Disclosure Statement

SAP enhancement package 4 for SAP ERP 6.0 (Unicode) Sales and Distribution Benchmark, results as of 9/19/2010. For more details, see http://www.sap.com/benchmark. SD-Parallel, Four Sun Fire X4470 (each 4 processors, 32 cores, 64 threads) 40,000 SAP SD Users, Cert# 2010039. SD-Parallel, Two Sun Fire X4470 (each 4 processors, 32 cores, 64 threads) 21,000 SAP SD Users, Cert# 2010029. SD 2-Tier, HP ProLiant DL980 G7 (8 processors, 64 cores, 128 threads) 18,180 SAP SD Users, Cert# 2010028. SD 2-Tier, Fujitsu PRIMEQUEST 1800E (8 processors, 64 cores, 128 threads) 16,000 SAP SD Users, Cert# 2010010. SD-Parallel, Four Sun Blade X6270 (each 2 processors, 8 cores, 16 threads) 13,718 SAP SD Users, Cert# 2009041. SD 2-Tier, HP ProLiant DL580 G7 (4 processors, 32 cores, 64 threads) 10,490 SAP SD Users, Cert# 2010032. SD 2-Tier, IBM System x3850 X5 (4 processors, 32 cores, 64 threads) 10,450 SAP SD Users, Cert# 2010012. SD 2-Tier, Fujitsu PRIMERGY RX600 S5 (4 processors, 32 cores, 64 threads) 9,560 SAP SD Users, Cert# 2010017. SD-Parallel, Two Sun Blade X6270 (each 2 processors, 8 cores, 16 threads) 7,220 SAP SD Users, Cert# 2009040. SD-Parallel, Sun Blade X6270 (2 processors, 8 cores, 16 threads) 3,800 SAP SD Users, Cert# 2009039. SD 2-Tier, Sun Fire X4270 (2 processors, 8 cores, 16 threads) 3,800 SAP SD Users, Cert# 2009033.

SAP ERP 6.0 (Unicode) Sales and Distribution Benchmark, results as of 9/19/2010. SD-Parallel, Five IBM System p 570 (each 8 processors, 16 cores, 32 threads) 37,040 SAP SD Users, Cert# 2008013.

Tuesday Jun 29, 2010

Sun Fire X2270 M2 Achieves Leading Single Node Results on ANSYS FLUENT Benchmark

Oracle's Sun Fire X2270 M2 server produced leading single node performance results running the ANSYS FLUENT benchmark cases as compared to the best single node results currently posted at the ANSYS FLUENT website. ANSYS FLUENT is a prominent MCAE application used for computational fluid dynamics (CFD).

  • The Sun Fire X2270 M2 server outperformed all single node systems in 5 of 6 test cases at the 12 core level, beating systems from Cray and SGI.
  • For the truck_14m test, the Sun Fire X2270 M2 server outperformed all single node systems at all posted core counts, beating systems from SGI, Cray and HP. When considering performance on a single node, the truck_14m model is most representative of customer CFD model sizes in the test suite.
  • The Sun Fire X2270 M2 server with 12 cores performed up to 1.3 times faster than the previous generation Sun Fire X2270 server with 8 cores.

Performance Landscape

Results are presented for six of the seven ANSYS FLUENT benchmark tests. The seventh test is not a practical test for a single system. Results are ratings, where bigger is better. A rating is the number of jobs that could be run in a single day (86,400 / run time). Competitive results are from the ANSYS FLUENT benchmark website as of 25 June 2010.

Single System Performance

ANSYS FLUENT Benchmark Tests
Results are Ratings, Bigger is Better
System Benchmark Test
eddy_417k turbo_500k aircraft_2m sedan_4m truck_14m truck_poly_14m
Sun Fire X2270 M2 1129.4 5391.6 1105.9 814.1 94.8 96.4
SGI Altix 8400EX 1338.0 5308.8 1073.3 796.3 - -
SGI Altix XE1300C 1299.2 5284.4 1071.3 801.3 90.2 -
Cray CX1 1060.8 5127.6 1069.6 808.6 86.1 87.5

Scaling of Benchmark Test truck_14m

ANSYS FLUENT truck_14m Model
Results are Ratings, Bigger is Better
System Cores Used
12 8 4 2 1
Sun Fire X2270 M2 94.8 73.6 41.4 21.0 10.4
SGI Altix XE1300C 90.2 60.9 41.1 20.7 9.0
Cray CX1 (X5570) - 71.7 33.2 18.9 8.1
HP BL460 G6 (X5570) - 70.3 38.6 19.6 9.2

Comparing System Generations, Sun Fire X2270 M2 to Sun Fire X2270

ANSYS FLUENT Benchmark Tests
Results are Ratings, Bigger is Better
System Benchmark Test
eddy_417k turbo_500k aircraft_2m sedan_4m truck_14m truck_poly_14m
Sun Fire X2270 M2 1129.4 5374.8 1103.8 814.1 94.8 96.4
Sun Fire X2270 981.5 4163.9 862.7 691.2 73.6 73.3

Ratio 1.15 1.29 1.28 1.18 1.29 1.32

Results and Configuration Summary

Hardware Configuration:

Sun Fire X2270 M2
2 x 2.93 GHz Intel Xeon X5670 processors
48 GB memory
1 x 500 GB 7200 rpm SATA internal HDD

Sun Fire X2270
2 x 2.93 GHz Intel Xeon X5570 processors
48 GB memory
2 x 24 GB internal striped SSDs

Software Configuration:

64-bit SUSE Linux Enterprise Server 10 SP 3 (SP 2 for X2270)
ANSYS FLUENT V12.1.2
ANSYS FLUENT Benchmark Test Suite

Benchmark Description

The following description is from the ANSYS FLUENT website:

The FLUENT benchmarks suite comprises of a set of test cases covering a large range of mesh sizes, physical models and solvers representing typical industry usage. The cases range in size from a few 100 thousand cells to more than 100 million cells. Both the segregated and coupled implicit solvers are included, as well as hexahedral, mixed and polyhedral cell cases. This broad coverage is expected to demonstrate the breadth of FLUENT performance on a variety of hardware platforms and test cases.

The performance of a CFD code will depend on several factors, including size and topology of the mesh, physical models, numerics and parallelization, compilers and optimization, in addition to performance characteristics of the hardware where the simulation is performed. The principal objective of this benchmark suite is to provide comprehensive and fair comparative information of the performance of FLUENT on available hardware platforms.

About the ANSYS FLUENT 12 Benchmark Test Suite

    CFD models tend to be very large where grid refinement is required to capture with accuracy conditions in the boundary layer region adjacent to the body over which flow is occurring. Fine grids are required to also determine accurate turbulence conditions. As such these models can run for many hours or even days as well using a large number of processors.

See Also

Disclosure Statement

All information on the FLUENT website (http://www.fluent.com) is Copyrighted 1995-2010 by ANSYS Inc. Results as of June 25, 2010.

Sun Fire X2270 M2 Demonstrates Outstanding Single Node Performance on MSC.Nastran Benchmarks

Oracle's Sun Fire X2270 M2 server results showed outstanding performance running the MCAE application MSC.Nastran as shown by the MD Nastran MDR3 serial and parallel test cases.

Performance Landscape

Complete information about the serial results presented below can be found on the MSC Nastran website.


MD Nastran MDR3 Serial Test Results
Platform Benchmark Problem
Results are total elapsed run time in seconds
xl0imf1 xx0xst0 xl1fn40 vl0sst1
Sun Fire X2270 M2 999 704 2337 115
Sun Blade X6275 1107 798 2285 120
Intel Nehalem 1235 971 2453 123
Intel Nehalem w/ SSD 1484 767 2456 120
IBM:P6 570 ( I8 )
1510 4612 132
IBM:P6 570 ( I4 ) 1016 1618 5534 147

Complete information about the parallel results presented below can be found on the MSC Nastran website.


MD Nastran MDR3 Parallel Test Results
Platform Benchmark Problem
Results are total elapsed run time in seconds
xx0cmd2 md0mdf1
Serial DMP=2 DMP=4 DMP=8 Serial DMP=2 DMP=4
Sun Blade X6275 840 532 391 279 880 422 223
Sun Fire X2270 M2 847 558 371 297 889 462 232
Intel Nehalem w/ 4 SSD 887 639 405
902 479 235
Intel Nehalem 915 561 408
922 470 251
IBM:P6 570 ( I8 ) 920 574 392 322


IBM:P6 570 ( I4 ) 959 616 419 343 911 469 242

Results and Configuration Summary

Hardware Configuration:

Sun Fire X2270 M2
2 x 2.93 GHz Intel Xeon X5670 processors
48 GB memory
4 x 24 GB SSDs (striped)

Software Configuration:

64-bit SUSE Linux Enterprise Server 10 SP 3
MSC Software MD 2008 R3
MD Nastran MDR3 benchmark test suite

Benchmark Description

The benchmark tests are representative of typical MSC.Nastran applications including both serial and parallel (DMP) runs involving linear statics, nonlinear statics, and natural frequency extraction as well as others. MD Nastran is an integrated simulation system with a broad set of multidiscipline analysis capabilities.

Key Points and Best Practices

  • The test cases for the MSC.Nastran module all have a substantial I/O component where 15% to 25% of the total run times are associated with I/O activity (primarily scratch files). The required scratch file size ranges from less than 1 GB on up to about 140 GB. To obtain best performance, it is important to have a high performance storage system when running MD Nastran.

  • To improve performance, it is possible to make use of the MD Nastran feature which sets the maximum amount of memory the application will use. This allows a user to configure where temporary files are held, including in memory file systems like tmpfs.

See Also

Disclosure Statement

MSC.Software is a registered trademark of MSC. All information on the MSC.Software website is copyrighted. MD Nastran MDR3 results from http://www.mscsoftware.com and this report as of June 28, 2010.

Sun Fire X2270 M2 Sets World Record on SPEC OMP2001 Benchmark

Oracle's Sun Fire X2270 M2 server running the Oracle Solaris 10 10/09 with the Oracle Solaris Studio 12 Update 1 compiler, produced the top x86 SPECompM2001 result for all 2-socket servers.

  • The Sun Fire X2270 M2 server with two Intel Xeon X5670 processors running 24 OpenMP threads achieved a SPEC OMP2001 result of 55,178 SPECompM2001.

  • The Sun Fire X2270 M2 server beat the Cisco B200 M2 system even thought the Cisco system used the faster Intel Xeon X5680 (3.33GHz) chips.

Performance Landscape

SPEC OMP2001 Performance Charts - bigger is better, selected results, please see www.spec.org for complete results. All results as of 06/28/10.

In the tables below
"Base" = SPECompMbase2001 and "Peak" = SPECompMpeak2001

SPEC OMPM2001 results

System Processors Base
Threads
Performance Results
Cores/
Chips
Type GHz Peak Base
Sun Fire X2270 M2 12/2 Xeon X5670 2.93 24 55178 49548
Cisco B200 M2 12/2 Xeon X5680 3.33 24 55072 52314
Intel SR1600UR 12/2 Xeon X5680 3.33 24 54249 51510
Intel SR1600UR 12/2 Xeon X5670 2.93 24 53313 50283

Results and Configuration Summary

Hardware Configuration:

Sun Fire X2270 M2
2 x 2.93 GHz Intel Xeon X5670
24 GB

Software Configuration:

Oracle Solaris 10 10/09
Oracle Solaris Studio 12 Update 1
SPEC OMP2001 suite v3.2

Benchmark Description

The SPEC OMPM2001 Benchmark Suite was released in June 2001 and tests HPC performance using OpenMP for parallelism.

  • 11 programs (3 in C and 8 in Fortran) parallelized using OpenMP API
Goals of the suite:
  • Targeted to mid-range (4-32 processor) parallel systems
  • Run rules, tools and reporting similar to SPEC CPU2006
  • Programs representative of HPC and Scientific Applications

The SPEC OMPL2001 Benchmark Suite was released in June 2001 and tests HPC performance using OpenMP for parallelism.

  • 9 programs (2 in C and 7 in Fortran) parallelized using OpenMP API
Goals of the suite:
  • Targeted to larger parallel systems
  • Run rules, tools and reporting similar to SPEC CPU2006
  • Programs representative of HPC and Scientific Applications

There are "base" variants of both the above metrics that require more conservative compilation, such as using the same flags for all benchmarks.

See Also

Disclosure Statement

SPEC, SPEComp reg tm of Standard Performance Evaluation Corporation. Results from www.spec.org as of 28 June 2010 and this report. Sun Fire X2270 M2 (2 chips, 12 cores, 24 OpenMP threads) 55,178 SPECompM2001;

Sun Fire X4170 M2 Sets World Record on SPEC CPU2006 Benchmark

Oracle's Sun Fire X4170 M2 server equipped with two Intel Xeon X5670 2.93 GHz processors and running the Oracle Solaris 10 operating system delivered the a world record score of 53.5 SPECfp_base2006.

  • The Sun Fire X4170 M2 server using the Oracle Solaris Studio Express 06/10 compiler delivered a world record result of 53.5 SPECfp_base2006.

  • The Sun Fire X4170 M2 server delivered 20% better performance on the SPECfp_base2006 benchmark compared to the IBM 780 POWER7 based system.

  • The Sun Fire X4170 M2 server beat systems from Supermicro (X8DTU-LN4F+), Dell (R710), IBM (x3650 M3) and Bull (R460 F2) on SPECfp_base2006.

Performance Landscape

SPEC CPU2006 Performance Charts - bigger is better, selected results, please see www.spec.org for complete results. All results as of 06/28/10.

In the tables below
"Base" = SPECint_base2006, SPECfp_base2006, SPECint_rate_base2006 or SPECfp_rate_base2006
"Peak" = SPECint2006, SPECfp2006, SPECint_rate2006 or SPECfp_rate2006

SPECfp2006 results

System Processors Performance Results
Cores/
Chips
Type GHz Peak Base
Sun Fire X4170 M2 12/2 Xeon X5670 2.93 57.6 53.5
Sun Fire X2270 M2 12/2 Xeon X5670 2.93 58.6 49.9
Supermicro X8DTU-LN4F+ 8/2 Xeon X5677 3.46 48.8 45.9
IBM x3650 M3 8/2 Xeon X5677 3.46 48.9 45.8
Bull R460 F2 8/2 Xeon X5677 3.46 49.3 45.8
Dell R710 8/2 Xeon X5677 3.46 49.3 45.8
Dell R710 12/2 Xeon X5680 3.33 48.5 45.0
IBM 780 16/2 POWER7 3.94 71.5 44.5
Dell R710 12/2 Xeon X5670 2.93 45.8 42.5

SPECint_rate2006 results

System Processors Base
Copies
Performance Results
Cores/
Chips
Type GHz Peak Base
Dell R815 24/2 Opteron 6176 2.3 24 401 314
Fijitsu BX922 S2 12/2 Xeon X5680 3.33 24 381 354
Dell R710 12/2 Xeon X5680 3.33 24 379 355
Sun Blade X6270 M2 12/2 Xeon X5680 3.33 24 369 337
Sun Fire X4170 M2 12/2 Xeon X5670 2.93 24 353 316
Sun Fire X2270 M2 (S10) 12/2 Xeon X5670 2.93 24 346 311
Sun Fire X2270 M2 (OEL) 12/2 Xeon X5670 2.93 24 342 320

SPECfp_rate2006 results

System Processors Base
Copies
Performance Results
Cores/
Chips
Type GHz Peak Base
Dell R815 24/2 Opteron 6176 2.3 24 323 295
Dell R710 12/2 Xeon X5680 3.33 24 256 248
Fijitsu BX922 S2 12/2 Xeon X5680 3.33 24 256 248
Sun Blade X6270 M2 12/2 Xeon X5680 3.33 24 255 247
Sun Fire X4170 M2 12/2 Xeon X5670 2.93 24 245 234
Sun Fire X2270 M2 (S10) 12/2 Xeon X5670 2.93 24 240 231
Sun Fire X2270 M2 (OEL) 12/2 Xeon X5670 2.93 24 235 226

Results and Configuration Summary

Hardware Configuration:

Sun Fire X4170
2 x 2.93 GHz Intel Xeon X5670
48 GB
Sun Fire X2270
2 x 2.93 GHz Intel Xeon X5670
48 GB
Sun Blade X6270
2 x 3.33 GHz Intel Xeon X5680
48 GB

Software Configuration:

Oracle Solaris 10 10/09
Oracle Solaris Studio Express 6/10
SPEC CPU2006 suite v1.1
MicroQuill SmartHeap Library v8.1

Benchmark Description

SPEC CPU2006 is SPEC's most popular benchmark, with over 8000 results published in the three years since it was introduced. It measures:

  • "Speed" - single copy performance of chip, memory, compiler
  • "Rate" - multiple copy (throughput)

The rate metrics are used for the throughput-oriented systems described on this page. These metrics include:

  • SPECint_rate2006: throughput for 12 integer benchmarks derived from real applications such as perl, gcc, XML processing, and pathfinding
  • SPECfp_rate2006: throughput for 17 floating point benchmarks derived from real applications, including chemistry, physics, genetics, and weather.

There are base variants of both the above metrics that require more conservative compilation. In particular, all benchmarks of a particular programming language must use the same compilation flags.

See Also

Disclosure Statement

SPEC, SPECint, SPECfp reg tm of Standard Performance Evaluation Corporation. Results from www.spec.org as of 24 June 2010 and this report. Sun Fire X4170 M2 53.5 SPECfp_base2006.

Sun Blade X6270 M2 Sets World Record on SPECjbb2005 Benchmark

Oracle's Sun Blade X6270 M2, equipped with two 3.33 Ghz Intel Xeon X5680 processors obtained a result of 931637 SPECjbb2005 bops, 465819 SPECjbb2005 bops/JVM on the SPECjbb2005 benchmark. This is a World Record result on 2-socket servers with Intel Xeon 5600 series x86 processors.

  • This result was obtained on the Sun Blade X6270 M2 using Microsoft Windows 2008 R2 and Java HotSpot(TM) 64-Bit Server VM on Windows, version 1.6.0_21 Performance Release.

Performance Landscape

SPECjbb2005 Performance Chart (ordered by performance)

bops: SPECjbb2005 Business Operations per Second (bigger is better)

System Processor JVM Performance
SPECjbb2005
bops
SPECjbb2005
bops/JVM
Sun Blade X6270 M2 Intel X5680 3.33 GHz Java HotSpot 1.6.0_21 931637 465819
Cisco UCS B200 M2 Intel X5680 3.33 GHz IBM J9 VM 2.4 931076 155179
Fujitsu TX300 S6 Intel X5680 3.33 GHz IBM J9 VM 2.4 928393 154732
IBM x3500 M3 Intel X5680 3.33 GHz IBM J9 VM 2.4 916251 152709

Complete benchmark results may be found at the SPEC benchmark website http://www.spec.org.

Results and Configuration Summary

Hardware Configuration:

Sun Blade X6270 M2
2 x Intel Xeon X5680 3.33 GHz processors
48 GB of memory

Software Configuration:

Windows Server 2008 R2 Enterprise Edition
Java HotSpot(TM) 64-Bit Server VM on Windows, version 1.6.0_21 Performance Release

Benchmark Description

SPECjbb2005 (Java Business Benchmark) measures the performance of a Java implemented application tier (server-side Java). The benchmark is based on the order processing in a wholesale supplier application. The performance of the user tier and the database tier are not measured in this test. The metrics given are number of SPECjbb2005 bops (Business Operations per Second) and SPECjbb2005 bops/JVM (bops per JVM instance).

Key Points and Best Practices

  • Enhancements to the JVM had a major impact on performance.

See Also

Disclosure Statement

SPEC, SPECjbb reg tm of Standard Performance Evaluation Corporation. Results as of 6/28/2010 on www.spec.org. Sun Blade X6270 M2(2 chips, 12 cores) 931637 SPECjbb2005 bops, 465819 SPECjbb2005 bops/JVM submitted for review. Cisco UCS B200 M2(2 chips, 12 cores) 931076 SPECjbb2005 bops, 155179 SPECjbb2005 bops/JVM. Fujitsu TX300 S6(2 chips, 12 cores) 928393 SPECjbb2005 bops, 154732 SPECjbb2005 bops/JVM. IBM x3500 M3(2 chips, 12 cores) 916251 SPECjbb2005 bops, 152709 SPECjbb2005 bops/JVM.

Monday Jun 28, 2010

Sun Fire X4270 M2 Sets World Record on SPECjbb2005 Benchmark

Oracle's Sun Fire X4270 M2, equipped with two 3.33 GHz Intel Xeon X5680 processors, obtained a result of 812,358 SPECjbb2005 bops, 812,358 SPECjbb2005 bops/JVM on the SPECjb2005 benchmark. This is the best x86 based system result using a single JVM.
  • This result was obtained on the Sun Fire X4270 M2 using Oracle Solaris 10 10/09 and Java HotSpot(TM) 64-Bit Server VM on Solaris, version 1.6.0_21 Performance Release.

Performance Landscape

SPECjbb2005 Performance Chart (ordered by performance)

bops: SPECjbb2005 Business Operations per Second (bigger is better)

System Processor JVM Performance
SPECjbb2005
bops
SPECjbb2005
bops/JVM
Sun Fire X4270 M2 Intel X5680 3.33 GHz Java HotSpot 1.6.0_21 812,358 812,358
HP DL380 G6 Intel X5570 2.93 GHz Java HotSpot 1.6.0_14 509,962 509,962
Sun Blade X6270 Intel X5570 2.93 GHz Java HotSpot 1.6.0_14 503,675 503,675

Complete benchmark results may be found at the SPEC benchmark website http://www.spec.org.

Results and Configuration Summary

Hardware Configuration:

Sun Fire X4270 M2
2 x Intel Xeon X5680, 3.33 GHz processors
48 GB of memory

Software Configuration:

Oracle Solaris 10 10/09
Java HotSpot(TM) 64-Bit Server VM on Solaris, version 1.6.0_21 Performance Release

Benchmark Description

SPECjbb2005 (Java Business Benchmark) measures the performance of a Java implemented application tier (server-side Java). The benchmark is based on the order processing in a wholesale supplier application. The performance of the user tier and the database tier are not measured in this test. The metrics given are number of SPECjbb2005 bops (Business Operations per Second) and SPECjbb2005 bops/JVM (bops per JVM instance).

Key Points and Best Practices

  • Enhancements to the JVM had a major impact on performance

See Also

Disclosure Statement

SPEC, SPECjbb reg tm of Standard Performance Evaluation Corporation. Results as of 6/28/2010 on www.spec.org. Sun Fire X4270 M2 812358 SPECjbb2005 bops, 812358 SPECjbb2005 bops/JVM submitted for review; HP DL380 G6, 509,962 SPECjbb2005 bops, 509,962 SPECjbb2005 bops/JVM; Sun Blade X6270, 503,675 SPECjbb2005, 503,675 SPECjbb2005 bops/JVM.

Sun Fire X4470 Sets World Records on SPEC OMP2001 Benchmarks

Oracle's Sun Fire X4470 server, with four Intel Xeon X7560 processors capable of running OpenMP applications with 64 compute threads, delivered outstanding performance on the both medium and large suites of the industry-standard SPEC OMP2001 benchmark.

  • The Sun Fire X4470 server running the Oracle Solaris 10 10/09 operating system with Oracle Solaris Studio 12 Update 1 compiler software, produced the top x86 result on SPECompM2001.

  • The Sun Fire X4470 server running the Oracle Solaris 10 10/09 operating system with Oracle Solaris Studio 12 Update 1 compiler software, produced the top x86 result on SPECompL2001.

  • The Sun Fire X4470 server beats IBM Power 750 Express POWER7 3.55 GHz SPECompM2001 score by 14%, while the Sun Fire X4470 server uses half the number of OpenMP threads compared to the IBM Power 750.
  • The Sun Fire X4470 server with four Intel Xeon 7560 processors, running 64 OpenMP threads, achieved SPEC OMP2001 results of 118,264 SPECompM2001 and 642,479 SPECompL2001.

  • The Sun Fire X4470 server produced better SPECompL2001 results than Cisco (UCS C460 M1) and Intel (QSSC-S4R) even though they all used the same number of Intel Xeon X7560 processors.

  • The Sun Fire X4470 server produced better SPECompM2001 results than Cisco (UCS C460 M1), SGI (Altix UV 10) and Intel (QSSC-S4R) even though they all used the same number of Intel Xeon X7560 processors.

Performance Landscape

SPEC OMP2001 Performance Charts - bigger is better, selected results, please see www.spec.org for complete results. All results as of 06/28/10.

In the tables below
"Base" = SPECompLbase2001 or SPECompMbase2001
"Peak" = SPECompLpeak2001 or SPECompMpeak2001

SPEC OMPL2001 results

System Processors Base
Threads
Performance Results
Cores/
Chips
Type GHz Peak Base
Sun Fire X4470 32/4 Xeon 7560 2.26 64 642479 615790
Cisco UCS C460 M1 32/4 Xeon 7560 2.26 64 628126 607818
Intel QSSC-S4R X7560 32/4 Xeon 7560 2.26 64 610386 591375
Sun/Fujitsu SPARC M8000 64/16 SPARC64 VII 2.52 64 581807 532576

SPEC OMPM2001 results

System Processors Base
Threads
Performance Results
Cores/
Chips
Type GHz Peak Base
Sun Fire X4470 32/4 Xeon 7560 2.26 64 118264 95650
Cisco UCS C460 M1 32/4 Xeon 7560 2.26 64 109077 100258
SGI Altix UV 10 32/4 Xeon 7560 2.26 64 107248 96797
Intel QSSC-S4R X7560 32/4 Xeon 7560 2.26 64 106369 98288
Sun/Fujitsu SPARC M8000 64/16 SPARC64 VII 2.52 64 104714 75418
IBM Power 750 Express 32/4 POWER7 3.55 128 104175 92957

Results and Configuration Summary

Hardware Configuration:

Sun Fire X4470
4 x 2.26 GHz Intel Xeon X7560
256 GB

Software Configuration:

Oracle Solaris 10 10/09
Oracle Solaris Studio Update 1
SPEC OMP2001 suite v3.2

Benchmark Description

The SPEC OMPM2001 Benchmark Suite was released in June 2001 and tests HPC performance using OpenMP for parallelism.

  • 11 programs (3 in C and 8 in Fortran) parallelized using OpenMP API
Goals of the suite:
  • Targeted to mid-range (4-32 processor) parallel systems
  • Run rules, tools and reporting similar to SPEC CPU2006
  • Programs representative of HPC and Scientific Applications

The SPEC OMPL2001 Benchmark Suite was released in June 2001 and tests HPC performance using OpenMP for parallelism.

  • 9 programs (2 in C and 7 in Fortran) parallelized using OpenMP API
Goals of the suite:
  • Targeted to larger parallel systems
  • Run rules, tools and reporting similar to SPEC CPU2006
  • Programs representative of HPC and Scientific Applications

There are "base" variants of both the above metrics that require more conservative compilation, such as using the same flags for all benchmarks.

See Also

Disclosure Statement

SPEC, SPEComp reg tm of Standard Performance Evaluation Corporation. Results from www.spec.org as of 28 June 2010 and this report. Sun Fire X4470 (4 chips, 32 cores, 64 OpenMP threads) 642,479 SPECompL2001, 118264 SPECompM2001.

Sun Fire X4470 Sets World Record on SPEC CPU2006 Rate Benchmark

Oracle's Sun Fire X4470 server delivered a world record SPECint_rate2006 result for all x86 systems with 4 chips.

  • The Sun Fire X4470 server with four Intel Xeon X7560 processors achieved a SPECint_rate2006 score of 788 and a SPECfp_rate2006 score of 573

  • The Sun Fire X4470 server delivered better 4 socket x86 system performance on the SPECint_rate2006 benchmark compared to HP (DL585 G7), Cicso (UCS C460 M1), Dell (R815) and IBM (x3850 X5).

  • The Sun Fire X4470 server delivered better performance on the SPECfp_rate2006 benchmark compared to similar Intel Xeon X7560 processor based systems from Cisco (UCS C460 M1), IBM (x3850 X5), and Fujitsu (RX600 S5).

Performance Landscape

SPEC CPU2006 Performance Charts - bigger is better, selected results, please see www.spec.org for complete results. All results as of 06/28/10.

In the tables below
"Base" = SPECint_rate_base2006 or SPECfp_rate_base2006
"Peak" = SPECint_rate2006 or SPECfp_rate2006

SPECint_rate2006 results

System Processors Base
Copies
Performance Results
Cores/
Chips
Type GHz Peak Base
Sun Fire X4470 32/4 Xeon X7560 2.26 64 788 724
HP DL585 G7 48/4 Opteron 6176 2.3 48 782 610
Cisco UCS C460 M1 32/4 Xeon X7560 2.26 64 772 723
Dell R815 48/4 Opteron 6174 2.20 48 771 602
IBM x3850 X5 32/4 Xeon X7560 2.26 64 770 720
Sun Fire X4640 48/8 Opteron 8435 2.6 48 730 574

SPECfp_rate2006 results

System Processors Base
Copies
Performance Results
Cores/
Chips
Type GHz Peak Base
Dell R815 48/4 Opteron 6174 2.20 48 626 574
HP DL585 G7 48/4 Opteron 6176 2.3 48 619 572
Sun Fire X4470 32/4 Xeon X7560 2.26 64 573 547
Cicso UCS C460 M1 32/4 Xeon X7560 2.26 64 568 549
IBM x3850 X5 32/4 Xeon X7560 2.26 64 560 543
Fujitsu RX600 S5 32/4 Xeon X7560 2.26 64 559 538
Sun Fire X4640 48/8 Opteron 8435 2.6 48 470 434

Results and Configuration Summary

Hardware Configuration:

Sun Fire X4470
4 x 2.26 GHz Intel Xeon X7560
256 GB

Software Configuration:

Oracle Solaris 10 10/09
Oracle Solaris Studio Express 6/10
SPEC CPU2006 suite v1.1
MicroQuill SmartHeap Library v8.1

Benchmark Description

SPEC CPU2006 is SPEC's most popular benchmark, with over 8000 results published in the three years since it was introduced. It measures:

  • "Speed" - single copy performance of chip, memory, compiler
  • "Rate" - multiple copy (throughput)

The rate metrics are used for the throughput-oriented systems described on this page. These metrics include:

  • SPECint_rate2006: throughput for 12 integer benchmarks derived from real applications such as perl, gcc, XML processing, and pathfinding
  • SPECfp_rate2006: throughput for 17 floating point benchmarks derived from real applications, including chemistry, physics, genetics, and weather.

There are base variants of both the above metrics that require more conservative compilation. In particular, all benchmarks of a particular programming language must use the same compilation flags.

See Also

Disclosure Statement

SPEC, SPECint, SPECfp reg tm of Standard Performance Evaluation Corporation. Results from www.spec.org as of 24 June 2010 and this report. Sun Fire X4470 788 SPECint_rate2006.

Sun Fire X4470 2-Node Configuration Sets World Record for SAP SD-Parallel Benchmark

Using two of Oracle's Sun Fire X4470 servers to run the SAP Enhancement Package 4 for SAP ERP 6.0 (Unicode) Sales and Distribution – Parallel (SD-Parallel) standard application benchmark, Oracle delivered a world record result. This was run using Oracle Solaris 10 and Oracle 11g Real Application Clusters (RAC) software.

  • The Sun Fire X4470 servers result of 21,000 users delivered more than twice the performance of the IBM System x3850 X5 system result of 10,450 users.

  • The Sun Fire X4470 servers result of 21,000 users beat the HP ProLiant DL980 G7 system result of 18,180 users. Both solutions used 8 Intel Xeon X7560 processors.

  • The Sun Fire X4470 servers result of 21,000 users beat the Fujitsu PRIMEQUEST 1800E system result of 16,000 users. Both solutions used 8 Intel Xeon X7560 processors.

  • This result shows how a compete software and hardware solution from Oracle, using Oracle RAC, Oracle Solaris and along with Oracle's Sun servers, can provide a superior performing solution when compared to the competition.

Performance Landscape

SAP Enhancement Package 4 for SAP ERP 6.0 (Unicode) Sales and Distribution Benchmark, select results presented in decreasing performance order. Both Parallel and 2-Tier solution results are listed in the table.

System OS
Database
Users SAPS Type Date
Two Sun Fire X4470
4xIntel Xeon X7560 @2.26GHz
256 GB
Solaris 10
Oracle 11g Real Application Clusters
21,000 115,300 Parallel 28-Jun-10
HP DL980 G7
8xIntel Xeon X7560 @2.26GHz
512 GB
Win Server 2008 R2 DE
SQL Server 2008
18,180 99,320 2-Tier 21-Jun-10
Fujitsu PRIMEQUEST 1800E
8xIntel Xeon X7560 @2.26GHz
512 GB
Win Server 2008 R2 DE
SQL Server 2008
16,000 87,550 2-Tier 30-Mar-10
Four Sun Blade X6270
2xIntel Xeon X5570 @2.93GHz
48 GB
Solaris 10
Oracle 10g Real Application Clusters
13,718 75,762 Parallel 12-Oct-09
IBM System x3850 X5
4xIntel Xeon X7560 @2.26GHz
256 GB
Win Server 2008 EE
DB2 9.7
10,450 57,120 2-Tier 30-Mar-10
HP DL580 G7
4xIntel Xeon X7560 @2.26GHz
256 GB
Win Server 2008 R2 DE
SQL Server 2008
10,445 57,020 2-Tier 21-Jun-10
Fujitsu PRIMERGY RX600 S5
4xIntel Xeon X7560 @2.26GHz
512 GB
Win Server 2008 R2 DE
SQL Server 2008
9,560 52,300 2-Tier 06-May-10
Two Sun Blade X6270
2xIntel Xeon X5570 @2.93GHz
48 GB
Solaris 10
Oracle 10g Real Application Clusters
7,220 39,420 Parallel 12-Oct-09
One Sun Blade X6270
2xIntel Xeon X5570 @2.93GHz
48 GB
Solaris 10
Oracle 10g Real Application Clusters
3,800 20,750 Parallel 12-Oct-09
Sun Fire X4270
2xIntel Xeon X5570 @2.93GHz
48 GB
Solaris 10
Oracle 10g
3,800 21,000 2-Tier 21-Aug-09

Complete benchmark results may be found at the SAP benchmark website http://www.sap.com/solutions/benchmark/sd.epx.

Results and Configuration Summary

Hardware Configuration:

2 x Sun Fire X4470 servers, each with
4 x Intel Xeon X7560 2.26 GHz (4 chips, 32 cores, 64 threads)
256 GB memory

Software Configuration:

Oracle 11g Real Application Clusters (RAC)
Oracle Solaris 10

Results Summary:

Number of SAP SD benchmark users:
21,000
Average dialog response time:
0.93 seconds
Throughput:

Dialog steps/hour:
6,918,000

SAPS:
115,300
SAP Certification:
2010029

Benchmark Description

The SAP Standard Application Sales and Distribution - Parallel (SD-Parallel) Benchmark is a two-tier ERP business test that is indicative of full business workloads of complete order processing and invoice processing, and demonstrates the ability to run both the application and database software on a single system. The SAP Standard Application SD Benchmark represents the critical tasks performed in real-world ERP business environments.

The SD-Parallel Benchmark consists of the same transactions and user interaction steps as the SD Benchmark. This means that the SD-Parallel Benchmark runs the same business processes as the SD Benchmark. The difference between the benchmarks is the technical data distribution.

An additional rule for parallel and distributed databases is one must equally distribute the benchmark users across all database nodes for the used benchmark clients (round-robin-method). Following this rule, all database nodes work on data of all clients. This avoids unrealistic configurations such as having only one client per database node.

The SAP Benchmark Council agreed to give the parallel benchmark a different name so that the difference can be easily recognized by any interested parties - customers, prospects, and analysts. The naming convention is SD-Parallel for Sales & Distribution - Parallel.

SAP is one of the premier world-wide ERP application providers, and maintains a suite of benchmark tests to demonstrate the performance of competitive systems on the various SAP products.

See Also

Disclosure Statement

SAP Enhancement Package 4 for SAP ERP 6.0 (Unicode) Sales and Distribution Benchmark, results as of 6/22/2010. For more details, see http://www.sap.com/benchmark. SD-Parallel, Two Sun Fire X4470 (each 4 processors, 32 cores, 64 threads) 21,000 SAP SD Users, Cert# 2010029. SD 2-Tier, HP ProLiant DL980 G7 (8 processors, 64 cores, 128 threads) 18,180 SAP SD Users, Cert# 2010028. SD 2-Tier, Fujitsu PRIMEQUEST 1800E (8 processors, 64 cores, 128 threads) 16,00o SAP SD Users, Cert# 2010010. SD-Parallel, Four Sun Blade X6270 (each 2 processors, 8 cores, 16 threads) 13,718 SAP SD Users, Cert# 2009041. SD 2-Tier, IBM System x3850 X5 (4 processors, 32 cores, 64 threads) 10,450 SAP SD Users, Cert# 2010012. SD 2-Tier, Fujitsu PRIMERGY RX600 S5 (4 processors, 32 cores, 64 threads) 9,560 SAP SD Users, Cert# 2010017. SD-Parallel, Two Sun Blade X6270 (each 2 processors, 8 cores, 16 threads) 7,220 SAP SD Users, Cert# 2009040. SD-Parallel, Sun Blade X6270 (2 processors, 8 cores, 16 threads) 3,800 SAP SD Users, Cert# 2009039. SD 2-Tier, Sun Fire X4270 (2 processors, 8 cores, 16 threads) 3,800 SAP SD Users, Cert# 2009033.

Sun Fire X4800 Sets World Record on SPECjbb2005 Benchmark

Oracle's Sun Fire X4800, equipped with eight 2.26 GHz Intel Xeon X7560 processors, obtained a result of 3369694 SPECjbb2005 bops, 421212 SPECjbb2005 bops/JVM on the SPECjb2005 benchmark. This is the best result for 8-socket servers with Intel 7500 series x86 processors.
  • This result was obtained on the Sun Fire X4800 using Oracle Solaris 10 10/09 and Java HotSpot(TM) 64-Bit Server VM on Solaris, version 1.6.0_21 Performance Release.

Performance Landscape

SPECjbb2005 Performance Chart (ordered by performance)

bops: SPECjbb2005 Business Operations per Second (bigger is better)

System Processor JVM Performance
SPECjbb2005
bops
SPECjbb2005
bops/JVM
Sun Fire X4800 Intel X7560 2.26 GHz Java HotSpot 1.6.0_21 3369694 421212
NEC 5800 Intel X7560 2.26 GHz Oracle JRockit(R) 6 P28.0.0 3343714 417964
Fujitsu 1800E Intel X7560 2.26 GHz Oracle JRockit(R) 6 P28.0.0 3321826 415228

Complete benchmark results may be found at the SPEC benchmark website http://www.spec.org.

Results and Configuration Summary

Hardware Configuration:

Sun Fire X4800
8 x Intel Xeon X7560, 2.26 GHz processors
512 GB of memory

Software Configuration:

Oracle Solaris 10 10/09
Java HotSpot(TM) 64-Bit Server VM on Solaris, version 1.6.0_21 Performance Release

Benchmark Description

SPECjbb2005 (Java Business Benchmark) measures the performance of a Java implemented application tier (server-side Java). The benchmark is based on the order processing in a wholesale supplier application. The performance of the user tier and the database tier are not measured in this test. The metrics given are number of SPECjbb2005 bops (Business Operations per Second) and SPECjbb2005 bops/JVM (bops per JVM instance).

Key Points and Best Practices

  • Enhancements to the JVM had a major impact on performance

See Also

Disclosure Statement

SPEC, SPECjbb reg tm of Standard Performance Evaluation Corporation. Results as of 6/28/2010 on www.spec.org. Sun Fire X4800(8 chips, 64 cores) 3369694 SPECjbb2005 bops, 421211 SPECjbb2005 bops/JVM submitted for review. NEC 5800(8 chips, 64 cores) 3343714 SPECjbb2005 bops, 417964 SPECjbb2005 bops/JVM. Fujitsu 1800E(8 chips, 64 cores) 3321826 SPECjbb2005 bops, 415228 SPECjbb2005 bops/JVM.

Sun Fire X4800 Sets World Records on SPEC CPU2006 Rate Benchmarks

Oracle's Sun Fire X4800 server combines the compute capacity of up to eight Intel Xeon X7560 processors with Oracle Solaris 10 and Oracle Solaris Studio Express software in a compact 5U form factor to provide better performance than any other x86 8-chip server, setting world records for both integer and floating point throughput on the SPEC CPU2006 benchmark for 8-chip x86 systems.
  • The Sun Fire X4800 server, in combination with Oracle Solaris 10 and Oracle Solaris Studio Express 6/10 compiler, delivered results of 1380 SPECint_rate2006 and 936 SPECfp_rate2006.

  • The combination of the Sun Fire X4800, Oracle Solaris and Oracle Solaris Studio compiler software resulted in a better score than Fujitsu's PRIMEQUEST 1800E server, running Red Hat Linux and using Intel compiler software.

Performance Landscape

SPEC CPU2006 Performance Charts - bigger is better, selected results, please see www.spec.org for complete results. All results as of 06/28/10.

In the tables below
"Base" = SPECint_rate_base2006 or SPECfp_rate_base2006
"Peak" = SPECint_rate2006 or SPECfp_rate2006

SPECint_rate2006 results

System Processors Base
Copies
Performance Results Comments
Cores/
Chips
Type GHz Peak Base
Sun Fire X4800 64/8 Xeon X7560 2.26 128 1380 1260 Solaris 10
Fujitsu 1800E 64/8 Xeon X7560 2.26 128 1339 1254 RHEL5.4 Linux
HP DL785 G6 48/8 Opteron 8439 2.8 48 800 629
HP DL585 G7 48/4 Opteron 6176 2.3 48 782 610
Sun Fire X4640 48/8 Opteron 8435 2.6 48 730 574
HP BL890c i2 32/8 Itanium 9350 1.73 32 531 507

SPECfp_rate2006 results

System Processors Base
Copies
Performance Results Comments
Cores/
Chips
Type GHz Peak Base
Sun Fire X4800 64/8 Xeon X7560 2.26 128 936 888 Solaris 10
Fujitsu 1800E 64/8 Xeon X7560 2.26 128 904 887 RHEL5.4 Linux
HP DL585 G7 48/4 Opteron 6176 2.3 48 619 572
HP BL890c i2 32/8 Itanium 9350 1.73 32 520 508
HP DL785 G6 48/8 Opteron 8439 2.8 48 513 473
Sun Fire X4640 48/8 Opteron 8435 2.6 48 470 434

Results and Configuration Summary

Hardware Configuration:

Sun Fire X4800
8 x 2.26 GHz Intel Xeon X7560
256 GB

Software Configuration:

Oracle Solaris 10 10/09
Oracle Solaris Studio Express 6/10
SPEC CPU2006 suite v1.1
MicroQuill SmartHeap Library v8.1

Benchmark Description

SPEC CPU2006 is SPEC's most popular benchmark, with over 8000 results published in the three years since it was introduced. It measures:

  • "Speed" - single copy performance of chip, memory, compiler
  • "Rate" - multiple copy (throughput)

The rate metrics are used for the throughput-oriented systems described on this page. These metrics include:

  • SPECint_rate2006: throughput for 12 integer benchmarks derived from real applications such as perl, gcc, XML processing, and pathfinding
  • SPECfp_rate2006: throughput for 17 floating point benchmarks derived from real applications, including chemistry, physics, genetics, and weather.

There are base variants of both the above metrics that require more conservative compilation. In particular, all benchmarks of a particular programming language must use the same compilation flags.

See Also

Disclosure Statement

SPEC, SPECint, SPECfp reg tm of Standard Performance Evaluation Corporation. Results from www.spec.org as of 24 June 2010 and this report. Sun Fire X4800 1380 SPECint_rate2006, 936 SPECfp_rate2006.

Tuesday Apr 06, 2010

WRF Benchmark: X6275 Beats Power6

Significance of Results

Oracle's Sun Blade X6275 cluster is 28% faster than the IBM POWER6 cluster on Weather Research and Forecasting (WRF) continental United Status (CONUS) benchmark datasets. The Sun Blade X6275 cluster used a Quad Data Rate (QDR) InfiniBand connection along with Intel compilers and MPI.

  • On the 12 km CONUS data set, the Sun Blade X6275 cluster was 28% faster than the IBM POWER6 cluster at 512 cores.

  • The Sun Blade X6275 cluster with 768 cores (one full Sun Blade 6048 chassis) was 47% faster than 1024 cores of the IBM POWER6 cluster (multiple racks).

  • On the 2.5 km CONUS data set, the Sun Blade X6275 cluster was 21% faster than the IBM POWER6 cluster at 512 cores.

  • The Sun Blade X6275 cluster with 768 cores (on full Sun Blade 6048 chassis) outperforms the IBM Power6 cluster with 1024 cores by 28% on the 2.5 km CONUS dataset.

Performance Landscape

The performance in GFLOPS is shown below on multiple datasets.

Weather Research and Forecasting
CONUS 12 KM Dataset
Cores Performance in GFLOPS
Sun
X6275
Intel
Whitebox
IBM
POWER6
Cray
XT5
SGI TACC
Ranger
Blue
Gene/P
8 17.5 19.8 17.0
10.2

16 38.7 37.5 33.6 21.4 20.1 10.8
32 71.6 73.3 66.5 40.4 39.8 21.2 5.9
64 132.5 131.4 117.2 75.2 77.0 37.8
128 235.8 232.8 209.1 137.4 114.0 74.5 20.4
192 323.6





256 405.2 415.1 363.1 243.2 197.9 121.0 37.4
384 556.6





512 691.9 696.7 542.2 392.2 375.2 193.9 65.6
768 912.0






1024

618.5 634.1 605.9 271.7 108.5
1700



840.1


2048





175.6

All cores used on each node which participates in each run.

Sun X6275 - 2.93 GHz X5570, InfiniBand
Intel Whitebox - 2.8 GHz GHz X5560, InfiniBand
IBM POWER6 - IBM Power 575, 4.7 GHz POWER6, InfiniBand, 3 frames
Cray XT5 - 2.7 GHz AMD Opteron (Shanghai), Cray SeaStar 2.1
SGI - best of a variety of results
TACC Ranger - 2.3 GHz AMD Opteron (Barcelona), InfiniBand
Blue Gene/P - 850 MHz PowerPC 450, 3D-Torus (proprietary)

Weather Research and Forecasting
CONUS 2.5 KM Dataset
Cores Performance in GFLOPS
Sun
X6275
SGI
8200EX
Blue
Gene/L
IBM
POWER6
Cray
XT5
Intel
Whitebox
TACC
Ranger
16 35.2






32 69.6

64.3


64 140.2

130.9
147.8 24.5
128 278.9 89.9
242.5 152.1 290.6 87.7
192 400.5





256 514.8 179.6 8.3 431.3 306.3 535.0 145.3
384 735.1





512 973.5 339.9 16.5 804.4 566.2 1019.9 311.0
768 1367.7





1024
721.5 124.8 1067.3 1075.9 1911.4 413.4
2048
1389.5 241.2
1849.7 3251.1
2600




4320.6
3072
1918.7 350.5
2651.3

4096
2543.5 453.2
3288.7

6144
3057.3 642.3
4280.1

8192
3569.7 820.4
5140.4

18432

1238.0



Sun X6275 - 2.93 GHz X5570, InfiniBand
SGI 8200EX - 3.0 GHz E5472, InfiniBand
Blue Gene/L - 700 MHz PowerPC 440, 3D-Torus (proprietary)
IBM POWER6 - IBM Power 575, 4.7 GHz POWER6, InfiniBand, 3 frames
Cray XT5 - 2.4 GHz AMD Opteron (Shanghai), Cray SeaStar 2.1
Intel Whitebox - 2.8 GHz GHz X5560, InfiniBand
TACC Ranger - 2.3 GHz AMD Opteron (Barcelona), InfiniBand

Results and Configuration Summary

Hardware Configuration:

48 x Sun Blade X6275 server modules, 2 nodes per blade, each node with
2 Intel Xeon X5570 2.93 GHz processors, turbo enabled, ht disabled
24 GB memory
QDR InfiniBand

Software Configuration:

SUSE Linux Enterprise Server 10 SP2
Intel Compilers 11.1.059
Intel MPI 3.2.2
WRF 3.0.1.1
WRF 3.1.1
netCDF 4.0.1

Benchmark Description

The Weather Research and Forecasting (WRF) Model is a next-generation mesoscale numerical weather prediction system designed to serve both operational forecasting and atmospheric research needs. WRF is designed to be a flexible, state-of-the-art atmospheric simulation system that is portable and efficient on available parallel computing platforms. It features multiple dynamical cores, a 3-dimensional variational (3DVAR) data assimilation system, and a software architecture allowing for computational parallelism and system extensibility.

There are two fixed-size benchmark cases.

Single domain, medium size 12KM Continental US (CONUS-12K)

  • 425x300x35 cell volume
  • 48hr, 12km resolution dataset from Oct 24, 2001
  • Benchmark is a 3hr simulation for hrs 25-27 starting from a provided restart file
  • Iterations output at every 72 sec of simulation time, with the computation cost of each time step ~30 GFLOP

Single domain, large size 2.5KM Continental US (CONUS-2.5K)

  • 1501x1201x35 cell volume
  • 6hr, 2.5km resolution dataset from June 4, 2005
  • Benchmark is the final 3hr simulation for hrs 3-6 starting from a provided restart file; the benchmark may also be performed (but seldom reported) for the full 6hrs starting from a cold start
  • Iterations output at every 15 sec of simulation time, with the computation cost of each time step ~412 GFLOP

See Also

Disclosure Statement

WRF, see http://www.mmm.ucar.edu/wrf/WG2/bench/, results as of 3/8/2010.

Monday Mar 29, 2010

Sun Blade X6275/QDR IB/ Reverse Time Migration

Significance of Results

Oracle's Sun Blade X6275 cluster with a Lustre file system was used to demonstrate the performance potential of the system when running reverse time migration applications complete with I/O processing.

  • Reduced the Total Application run time for the Reverse Time Migration when processing 800 input traces for two production sized surveys from a QDR Infiniband Lustre file system on 24 X6275 nodes, by implementing algorithm I/O optimizations and taking advantage of MPI I/O features in HPC ClusterTools:

    • 1243x1151x1231 - Wall clock time reduced from 11.5 to 6.3 minutes (1.8x improvement)
    • 2486x1151x1231 - Wall clock time reduced from 21.5 to 13.5 minutes (1.6x improvement)
  • Reduced the I/O Intensive Trace-Input time for the Reverse Time Migration when reading 800 input traces for two production sized surveys from a QDR Infiniband Lustre file system on 24 X6275 nodes running HPC ClusterTools, by modifying the algorithm to minimize the per node data requirement and avoiding unneeded synchronization:

    • 2486x1151x1231 : Time reduced from 121.5 to 3.2 seconds (38.0x improvement)
    • 1243x1151x1231 : Time reduced from 71.5 to 2.2 seconds (32.5x improvement)
  • Reduced the I/O Intensive Grid Initialization time for the Reverse Time Migration Grid when reading the Velocity, Epsilon, and Delta slices for two production sized surveys from a QDR Infiniband Lustre file system on 24 X6275 nodes running HPC ClusterTools, by modifying the algorithm to minimize the per node grid data requirement:

    • 2486x1151x1231 : Time reduced from 15.6 to 4.9 seconds (3.2x improvement)
    • 1243x1151x1231 : Time reduced from 8.9 to 1.2 seconds (7.4x improvement)

Performance Landscape

In the tables below, the hyperthreading feature is enabled and the systems are fully utilized.

This first table presents the total application performance in minutes. The overall performance improved significantly because of the improved I/O performance and other benefits.


Total Application Performance Comparison
Reverse Time Migration - SMP Threads and MPI Mode
Nodes 1243 x 1151 x 1231
800 Samples
2486 x 1151 x 1231
800 Samples
Original
Time (mins)
MPI I/O
Time (mins)
Improvement Original
Time (mins)
MPI I/O
Time (mins)
Improvement
24 11.5 6.3 1.8x 21.5 13.5 1.6x
20 12.0 8.0 1.5x 21.9 15.4 1.4x
16 13.8 9.7 1.4x 26.2 18.0 1.5x
12 21.7 13.2 1.6x 29.5 23.1 1.3x

This next table presents the initialization I/O time. The results are presented in seconds and shows the advantage of the improved MPI I/O strategy.


Initialization Time Performance Comparison
Reverse Time Migration - SMP Threads and MPI Mode
Nodes 1243 x 1151 x 1231
800 Samples
2486 x 1151 x 1231
800 Samples
Original
Time (sec)
MPI I/O
Time (sec)
Improvement Original
Time (sec)
MPI I/O
Time (sec)
Improvement
24 8.9 1.2 7.4x 15.6 4.9 3.2x
20 9.3 1.5 6.2x 16.9 3.9 4.3x
16 9.7 2.5 3.9x 17.4 11.3 1.5x
12 9.8 3.3 3.0x 22.5 14.9 1.5x

This last table presents the trace I/O time. The results are presented in seconds and shows the significant advantage of the improved MPI I/O strategy.


Trace I/O Time Performance Comparison
Reverse Time Migration - SMP Threads and MPI Mode
Nodes 1243 x 1151 x 1231
800 Samples
2486 x 1151 x 1231
800 Samples
Original
Time (sec)
MPI I/O
Time (sec)
Improvement Original
Time (sec)
MPI I/O
Time (sec)
Improvement
24 71.5 2.2 32.5x 121.5 3.2 38.0x
20 67.7 2.4 28.2x 118.3 3.9 30.3x
16 64.2 2.7 23.7x 110.7 4.6 24.1x
12 69.9 4.2 16.6x 296.3 14.6 20.3x

Results and Configuration Summary

Hardware Configuration:

Oracle's Sun Blade 6048 Modular Modular System with
12 x Oracle's Sun Blade x6275 Server Modules, each with
4 x 2.93 GHz Intel Xeon QC X5570 processors
12 x 4 GB memory at 1333 MHz
2 x 24 GB Internal Flash
QDR InfiniBand Lustre 1.8.0.1 File System

Software Configuration:

OS: 64-bit SUSE Linux Enterprise Server SLES 10 SP 2
MPI: Oracle Message Passing Toolkit 8.2.1 for I/O optimization to Lustre file system
MPI: Scali MPI Connect 5.6.6-59413 for original Lustre file system runs
Compiler: Oracle Solaris Studio 12 C++, Fortran, OpenMP

Benchmark Description

The primary objective of this Reverse Time Migration Benchmark is to present MPI I/O tuning techniques, exploit the power of the Sun's HPC ClusterTools MPI I/O implementation, and demonstrate the world-class performance of Sun's Lustre File System to Exploration Geophysicists throughout the world. A Sun Blade 6048 Modular System with 12 Sun Blade X6275 server modules were clustered together with a QDR Infiniband Lustre File System to show performance improvements in the Reverse Time Migration Throughput by using the Sun HPC ClusterTools MPI-IO features to implement specific algorithm I/O optimizations.

This Reverse Time Migration Benchmark measures the total time it takes to image 800 samples of various production size grids and write the final image to disk. In this new I/O optimized version, each node reads in only the data to be processed by that node plus a 4 element inline pad shared with it's neighbors to the left and right. This latest version, essentially loads the boundary condition data during the initialization phase. The previous version handled boundary conditions by having each node read in all the trace, velocity, and conditioning data. Or, alternatively, the master node would read in all the data and distribute it in it's entirety to every node in the cluster. With the previous version, each node had full memory copies of all input data sets even when it only processed a subset of that data. The new version only holds the inline dimensions and pads to be processed by a particular node in memory.

Key Points and Best Practices

  • The original implementation of the trace I/O involved the master node reading in nx \* ny floats and communicating this trace data to all the other nodes in a synchronous manner. Each node only used a subset of the trace data for each of the 800 time steps. The optimized I/O version has each node asynchronously read in only the (nx/num_procs + 8) \* ny floats that it will be processing. The additional 8 inline values for the optimized I/O version are the 4 element pads of a node's left and right neighbors to handle initial boundary conditions. The MPI_Barrier needed for the original implementation, for synchronization, and the additional I/O for each node to load all the data values, truly impacts performance. For the I/O optimized version, each node reads only the data values it needs and does not require the same MPI_Barrier synchronization as the original version of the Reverse Time Migration Benchmark. By performing such I/O optimizations, a significant improvement is seen in the Trace I/O.

  • For the best MPI performance, allocate the X6275 nodes in blade by blade order and run with HyperThreading enabled. The "Binary Conditioning" part of the Reverse Time Migration specifically likes hyperthreading.

  • To get the best I/O performance, use a maximum of 70% of each nodes available memory for the Reverse Time Migration application. Execution time may vary I/O results can occur if the nodes have different memory size configurations.

See Also

Monday Nov 02, 2009

Sun Ultra 27 Delivers Leading Single Frame Buffer SPECviewperf 10 Results

A Sun Ultra 27 workstation configured with an nVidia FX5800 graphics card delivered outstanding performance running the SPECviewperf® 10 benchmark.

  • When compared with other workstations running a single graphics card (i.e. not running two or more cards in SLI mode), the Sun Ultra 27 workstation places first in 6 of 8 subtests and second in the remaining two subtests.

  • The calculated geometric mean shows that Sun Ultra 27 workstation is 11% faster than competitor's workstations.

  • The optimum point for price/performance is the nVidia FX1800 graphics card.

Results have been published on the SPEC web site at http://www.spec.org/gwpg/gpc.data/vp10/summary.html.

Performance Landscape

Performance of the Sun Ultra 27 versus the competition. Bigger is better for each of the eight tests. The comparison is based upon the performance of the Sun Ultra 27 workstation. Performance is measured in frames per second.


3DSMAX CATIA ENSIGHT MAYA
Perf % Perf % Perf % Perf %
Sun Ultra 27 FX5800 59.34
68.81
58.07
246.09
HP xw4600 ATI FireGL V7700 49.71 19 48.05 43 57.11 2
268.62 -8
HP xw4600 FX4800 52.26 14 63.26 12 53.79 8
226.82 7
Fujtsu Celsius M470 FX3800 53.67 11 65.25 7 52.19 10 227.37 7

PROENGINEER SOLIDWORKS TEAMCENTER UGS
Perf % Perf % Perf % Perf %
Sun Ultra 27 FX5800 68.96
152.01
42.02
36.04
HP xw4600 ATI FireGL V7700 47.25 32 109.71 28 40.18 4 56.65 -57
HP xw4600 FX4800 61.15 11 131.31 14 28.42 32 33.43 7
Fujtsu Celsius M470 FX3800 64.39 7
139.2 8 29.02 31 33.27 8

Comparison of various frame buffers on the Sun Ultra 27 running SPECviewperf 10. Performance is reported for each test along with the difference in performance as compared to the FX5800 frame buffer. The runs in the table below were made with 3.2GHz W3570 processors.


3DSMAX CATIA ENSIGHT MAYA PROENGR SOLIDWRKS TEAMCNTR UGS
Perf % Perf % Perf % Perf % Perf % Perf % Perf % Perf %
FX5800 57.07
67.84
58.63
219.4
68.05
152.3
40.85
34.73
FX3800 57.17 0 66.57 2
54.91 7
206.4 6 66.48 2 146.3 4 38.48 6 33.12 5
FX1800 56.73 1
64.33 6
52.05 13 189.3 16 64.67 5 135.2 13 34.18 20
30.46 14
FX380 45.90 24 55.81 22 34.93 68 120.3 82 46.09 48 64.11 138 17.00 140 13.88 150

Results and Configuration Summary

Hardware Configuration:

    Sun Ultra 27 Workstation
    1 x 3.33 GHz Intel Xeon (tm) W3580
    2GB (1 x 2GB PC10600 1333MHz)
    1 x 500GB SATA
    nVidia Quadro FX380, FX1800, FX3800 & FX5800
    $7,529.00 (includes Microsoft Windows and monitor)

Software Configuration:

    OS: Microsoft Windows Vista Ultimate, 32-bit
    Benchmark: SPECviewperf 10

Benchmark Description

SPECviewperf measures 3D graphics rendering performance of systems running under OpenGL. SPECviewperf is a synthetic benchmark designed to be a predictor of application performance and a measure of graphics subsystem performance. It is a measure of graphics subsystem performance (primarily graphics bus, driver and graphics hardware) and its impact on the system without the full overhead of an application. SPECviewperf reports performance in frames per second.

Please go here for a more complete description of the tests.

Key Points and Best Practices

SPECviewperf measures the 3D rendering performance of systems running under OpenGL.

The SPECopcSM project group's SPECviewperf 10 is totally new performance evaluation software. In addition to features found in previous versions, it now provides the ability to compare performance of systems running in higher-quality graphics modes that use full-scene anti-aliasing, and measures how effectively graphics subsystems scale when running multithreaded graphics content. Since the SPECviewperf source and binaries have been upgraded to support changes, no comparisons should be made between past results and current results for viewsets running under SPECviewperf 10.

SPECviewperf 10 requires OpenGL 1.5 and a minimum of 1GB of system memory. It currently supports Windows 32/64.

See Also

Disclosure Statement

SPEC® and the benchmark name SPECviewperf® are registered trademarks of the Standard Performance Evaluation Corporation. Competitive benchmark results stated above reflect results published on www.spec.org as of Oct 18, 2009. For the latest SPECviewperf benchmark results, visit www.spec.org/gwpg.

Saturday Oct 24, 2009

Sun F5100 and Seismic Reverse Time Migration with faster Optimal Checkpointing

A prominent Seismic Processing algorithm, Reverse Time Migration with Optimal Checkpointing, in SMP "THREADS" Mode, was testing using a Sun Fire X4270 server configured with four high performance 15K SAS hard disk drives (HDDs) and a Sun Storage F5100 Flash Array. This benchmark compares I/O devices for checkpointing wave state information while processing a production seismic migration.

  • Sun Storage F5100 Flash Array is 2.2x faster than high-performance 15K RPM disks.

  • Multithreading the checkpointing using the Sun Studio C++ Compiler OpenMP implementation gives a 12.8x performance improvement over the original single threaded version.

These results show the new trend in seismic processing to run iterative Reverse Time Migrations and migration playback is a reality. This is made possible through the use of Sun FlashFire technology to provide good checkpointing speeds without additional disk cache memory. The application can take advantage of all the memory within a node without regard to checkpoint cache buffers required for performance to HDDs. Similarly, larger problem sizes can be solved without increasing the memory footprint of each computational node.

Performance Landscape


Reverse Time Migration Optimal Checkpointing - SMP Threads Mode
Grid Size -800 x 1151 x 1231 with 800 Samples - 60GB of memory
Number
Checkpts
HDD F5100
Put Time
(secs)
Get Time
(secs)
Total Time
(secs)
Put Time
(secs)
Get Time
(secs)
Total Time
(secs)
F5100
Speedup
80 660.8 25.8 686.6 277.4 40.2 317.6 2.2x
400 1615.6 382.3 1997.9 989.5 269.7 1259.2 1.6x


Reverse Time Migration Optimal Checkpointing - SMP Threads Mode
Grid Size -125 x 1151 x 1231 with 800 Samples - 9GB of memory
Number
Checkpts
HDD F5100
Put Time
(secs)
Get Time
(secs)
Total Time
(secs)
Put Time
(secs)
Get Time
(secs)
Total Time
(secs)
F5100
Speedup
80 10.2 0.2 10.4 8.0 0.2 8.2 1.3x
400 52.3 0.4 52.7 45.2 0.3 45.5 1.2x
800 102.6 0.7 103.3 91.8 0.6 92.4 1.1x


Reverse Time Migration Optimal Checkpointing
Single Thread vs Multithreaded I/O Performance
Grid Size -125 x 1151 x 1231 with 800 Samples - 9GB of memory
Number
Checkpts
Single Thread F5100
Total Time (secs)
Multithreaded F5100
Total Time (secs)
Multithread
Speedup
80 105.3 8.2 12.8x
400 482.9 45.5 10.6x
800 963.5 92.4 10.4x

Note: Hyperthreading and Turbo Mode enabled while running 16 threads per node.

Results and Configuration Summary

Hardware Configuration:

    Sun Fire 4270 Server
      2 x 2.93 GHz Quad-core Intel Xeon X5570 processors
      72 GB memory
      4 x 73 GB 15K SAS drives
        File system striped across 4 15K RPM high-performance SAS HD RAID0
      Sun Storage F5100 Flash Array with local/internal r/w buff 4096
        20 x 24 GB flash modules

Software Configuration:

    OS: 64-bit SUSE Linux Enterprise Server SLES 10 SP 2
    Compiler: Sun Studio 12 C++, Fortran, OpenMP

Benchmark Description

The Reverse Time Migration (RTM) is currently the most popular seismic processing algorithm because of it's ability to produce quality images of complex substructures. It can accurately image steep dips that can not be imaged correctly with traditional Kirchhoff 3D or frequency domain algorithms. The Wave Equation Migration (WEM) can image steep dips but does not produce the image quality that can be achieved by the RTM. However, the increased computational complexity of the RTM over the WEM introduces new areas for performance optimization. The current trend in seismic processing is to perform iterative migrations on wide azimuth marine data surveys using the Reverse Time Migration.

The Reverse Time Migration with Optimal Checkpointing was introduced so large migrations could be performed within minimal memory configurations of x86 cluster nodes. The idea is to only have three wavestate vectors in memory for each of the source and receiver wavefields instead of holding the entire wavefields in memory for the duration of processing. With the Sun Flash F5100, this can be done with little performance penalty to the full migration time. Another advantage of checkpointing is to provide the ability to playback migrations and facilitate iterative migrations.

  • The stored snapshot data can be reprocessed with different filtering, image conditioning, or a variety of other parameters.
  • Fine grain snapshoting can help the processing of more complex subsurface data.
  • A Geoscientist can "playback" a migration from the saved snapshots to visually validate migration accuracy or pick areas of interest for additional processing.

The Reverse Time Migration with Optimal Checkpointing is an algorithm designed by Griewank (Griewank, 1992; Blanch et al., 1998; Griewank, 2000; Griewank and Walther, 2000; Akcelik et al., 2003).

  • The application takes snapshots of wavefield state data for some interval of the total number of samples.
  • This adjoint state method performs crosscorrelation of the source and receiver wavefields at the each level.
  • Forward recursion is used for the source wavefield and backward recursion for the receiver wavefield.
  • For relatively small seismic migrations, all of the forward processed state information can be saved and restored with minimal impact on the total processing time.
  • Effectively, the computational complexity increases while the memory requirements decrease by a logarithmic factor of the number of snapshots.
  • Griewank's algorithm helps define the most optimal tradeoff between computational performance and the number of memory buffers (memory requirements) to support this cross correlation.

For the purposes of this benchmark, this implementation of the Reverse Time Migration with Optimal Checkpointing does not fully implement the optimal memory buffer scheme proposed by Griewank. The intent is to compare various I/O alternatives for saving wave state data for each node in a compute cluster.

This benchmark measures the time to perform the wave state saves and restores while simultaneously processing the wave state data.

Key Points and Best Practices

  • Mulithreading the checkpointing using Sun Studio OpenMP and running 16 I/O threads with hyperthreading enabled gives a performance advantage over single threaded I/O to the Sun Storage F5100 flash array. The Sun Storage F5100 flash array can process concurrent I/O requests from multiple threads very efficiently.
  • Allocating the majority of a node's available memory to the Reverse Time Migration algorithm and leaving little memory for I/O caching favors the Sun Storage F5100 flash array over direct attached high performance disk drives. This performance advantage decreases as the number of snapshots increase. The reason for this is that increasing the number of snapshots decreases the memory requirement for the application.

See Also

Disclosure Statement

Reverse Time Migration with Optimal Checkpointing, Results as of 10/23/2009. For more info http://www.caam.rice.edu/tech_reports/2006/TR06-18.pdf

Sun C48 & Lustre fast for Seismic Reverse Time Migration using Sun X6275

Significance of Results

A Sun Blade 6048 Modular System with 12 Sun Blade X6275 server modules were clustered together with QDR InfiniBand and using a Lustre File System with QDR InfiniBand to show performance improvements over an NFS file system for reading in Velocity, Epsilon, and Delta Slices and imaging 800 samples of various various grid sizes using the Reverse Time Migration.

  • The Initialization Time for populating the processing grids demonstrates significant advantages of Lustre over NFS:
    • 2486x1151x1231 : 20x improvement
    • 1243x1151x1231 : 20x improvement
    • 125x1151x1231 : 11x improvement
  • The Total Application Performance shows the Interconnect and I/O advantages of using QDR InfiniBand Lustre for the large grid sizes:
    • 2486x1151x1231 : 2x improvement - processed in less than 19 minutes
    • 1243x1151x1231 : 2x improvement - processed in less than 10 minutes

  • The Computational Kernel Scalability Efficiency for the 3 grid sizes:
    • 125x1151x1231 : 97% (1-8 nodes)
    • 1243x1151x1231 : 102% (8-24 nodes)
    • 2486x1151x1231 : 100% (12-24 nodes)

  • The Total Application Scalability Efficiency for the large grid sizes:
    • 1243x1151x1231 : 72% (8-24 nodes)
    • 2485x1151x1231 : 71% (12-24 nodes)

  • On the X5570 Intel processor with HyperThreading enabled and running 16 OpenMP threads per node gives approximately a 10% performance improvement over running 8 threads per node.

Performance Landscape

This first table presents the initialization time, comparing different number processors along with different problem sizes. The results are presented in seconds and shows the advantage the Lustre file system running over QDR InfiniBand provided when compared to a simple NFS file system.


Initialization Time Performance Comparison
Reverse Time Migration - SMP Threads and MPI Mode
Nodes Procs 125 x 1151 x 1231
800 Samples
1243 x 1151 x 1231
800 Samples
2486 x 1151 x 1231
800 Samples
Lustre Time
(sec)
NFS Time
(sec)
Lustre Time
(sec)
NFS Time
(sec)
Lustre Time
(sec)
NFS Time
(sec)
24 48 1.59 18.90 8.90 181.78 15.63 362.48
20 40 1.60 18.90 8.93 181.49 16.91 358.81
16 32 1.58 18.59 8.97 181.58 17.39 353.72
12 24 1.54 18.61 9.35 182.31 22.50 364.25
8 16 1.40 18.60 10.02 183.79

4 8 1.57 18.80



2 4 2.54 19.31



1 2 4.54 20.34



This next table presents the total application run time, comparing different number processors along with different problem sizes. It shows that for larger problems, using the Lustre file system running over QDR InfiniBand provided a big performance advantage when compared to a simple NFS file system.


Total Application Performance Comparison
Reverse Time Migration - SMP Threads and MPI Mode
Nodes Procs 125 x 1151 x 1231
800 Samples
1243 x 1151 x 1231
800 Samples
2486 x 1151 x 1231
800 Samples
Lustre Time
(sec)
NFS Time
(sec)
Lustre Time
(sec)
NFS Time
(sec)
Lustre Time
(sec)
NFS Time
(sec)
24 48 251.48 273.79 553.75 1125.02 1107.66 2310.25
20 40 232.00 253.63 658.54 971.65 1143.47 2062.80
16 32 227.91 209.66 826.37 1003.81 1309.32 2348.60
12 24 217.77 234.61 884.27 1027.23 1579.95 3877.88
8 16 223.38 203.14 1200.71 1362.42

4 8 341.14 272.68



2 4 605.62 625.25



1 2 892.40 841.94



The following table presents the run time and speedup of just the computational kernel for different processor counts for the three different problem sizes considered. The scaling results are based upon the smallest number of nodes run and that number is used as the baseline reference point.


Computational Kernel Performance & Scalability
Reverse Time Migration - SMP Threads and MPI Mode
Nodes Procs 125 x 1151 x 1231
800 Samples
1243 x 1151 x 1231
800 Samples
2486 x 1151 x 1231
800 Samples
X6275 Time
(sec)
Speedup:
1-node
X6275 Time
(sec)
Speedup:
1-node
X6275 Time
(sec)
Speedup:
1-node
24 48 35.38 13.7 210.82 24.5 427.40 24.0
20 40 35.02 13.8 255.27 20.2 517.03 19.8
16 32 41.76 11.6 317.96 16.2 646.22 15.8
12 24 49.53 9.8 422.17 12.2 853.37 12.0\*
8 16 62.34 7.8 645.27 8.0\*

4 8 124.66 3.9



2 4 238.80 2.0



1 2 484.89 1.0



The last table presents the speedup of the total application for different processor counts for the three different problem sizes presented. The scaling results are based upon the smallest number of nodes run and that number is used as the baseline reference point.


Total Application Scalability Comparison
Reverse Time Migration - SMP Threads and MPI Mode
Nodes Procs 125 x 1151 x 1231
800 Samples
Lustre Speedup:
1-node
1243 x 1151 x 1231
800 Samples
Lustre Speedup:
1-node
2486 x 1151 x 1231
800 Samples
Lustre Speedup:
1-node
24 48 3.6 17.3 17.1
20 40 3.8 14.6 16.6
16 32 4.0 11.6 14.5
12 24 4.1 10.9 12.0\*
8 16 4.0 8.0\*
4 8 2.6

2 4 1.5

1 2 1.0

Note: HyperThreading is enabled and running 16 threads per Node.

Results and Configuration Summary

Hardware Configuration:
    Sun Blade 6048 Modular Modular System with
      12 x Sun Blade x6275 Server Modules, each with
        4 x 2.93 GHz Intel Xeon QC X5570 processors
        12 x 4 GB memory at 1333 MHz
        2 x 24 GB Internal Flash
    QDR InfiniBand Lustre 1.8.0.1 File System
    GBit NFS file system

Software Configuration:

    OS: 64-bit SUSE Linux Enterprise Server SLES 10 SP 2
    MPI: Scali MPI Connect 5.6.6-59413
    Compiler: Sun Studio 12 C++, Fortran, OpenMP

Benchmark Description

The Reverse Time Migration (RTM) is currently the most popular seismic processing algorithm because of its ability to produce quality images of complex substructures. It can accurately image steep dips that can not be imaged correctly with traditional Kirchhoff 3D or frequency domain algorithms. The Wave Equation Migration (WEM) can image steep dips but does not produce the image quality that can be achieved by the RTM. However, the increased computational complexity of the RTM over the WEM introduces new areas for performance optimization. The current trend in seismic processing is to perform iterative migrations on wide azimuth marine data surveys using the Reverse Time Migration.

This Reverse Time Migration code reads in processing parameters that define the grid dimensions, number of threads, number of processors, imaging condition, and various other parameters. The master node calculates the memory requirements to determine if there is sufficient memory to process the migration "in-core". The domain decomposition across all the nodes is determined by dividing the first grid dimension by the number of nodes. Each node then reads in it's section of the Velocity Slices, Delta Slices, and Epsilon Slices using MPI IO reads. The three source and receiver wavefield state vectors are created: previous, current, and next state. The processing steps through the input trace data reading both the receiver and source data for each of the 800 time steps. It uses forward propagation for the source wave field and backward propagation in time to cross correlate the receiver wavefield. The computational kernel consists of a 13 point stencil to process a subgrid within the memory of each node using OpenMP parallelism. Afterwards, conditioning and absorption are applied and boundary data is communicated to neighboring nodes as each time step is processed. The final image is written out using MPI IO.

Total memory requirements for each grid size:

    125x1151x1231: 7.5GB
    1243x1151x1231: 78GB
    2486x1151x1231: 156GB

For this phase of benchmarking, the focus was to optimize the data initialization. In the next phase of benchmarking, the trace data reading will be optimized so that each node reads in only it's section of interest. In this benchmark the trace data reading skews the Total Application Performance as the number of nodes increase. This will be optimized in the next phase of benchmarking, as well as, further node optimization with OpenMP. The IO description for this benchmark phase on each grid size:

    125x1151x1231:
      Initialization MPI Read: 3 x 709MB = 2.1GB / number of nodes
      Trace Data Read per Node: 2 x 800 x 576KB = 920MB \* number of nodes
      Final Output Image MPI Write: 709MB / number of nodes
    1243x1151x1231: 78GB
      Initialization MPI Read: 3 x 7.1GB = 21.3GB / number of nodes
      Trace Data Read per Node: 2 x 800 x 5.7MB = 9.2GB \* number of nodes
      Final Output Image MPI Write: 7.1GB / number of nodes
    2486x1151x1231: 156GB
      Initialization MPI Read: 3 x 14.2GB = 42.6GB / number of nodes
      Trace Data Read per Node: 2 x 800 x 11.4MB = 18.4GB \* number of nodes
      Final Output Image MPI Write: 42.6GB / number of nodes

Key Points and Best Practices

  • Additional evaluations were performed to compare GBit NFS, Infiniband NFS, and Infiniband Lustre for the Reverse Time Migration Initialization. Infiniband NFS was 6x faster than GBit NFS and Infiniband Lustre was 3x faster than Infiniband NFS using the same disk configurations. On 12 nodes for grid size 2486x1151x1231 the initialization time was 22.50 seconds for IB Lustre, 61.03 seconds for IB NFS, and 364.25 seconds for GBit NFS.
  • The Reverse Time Migration computational performance scales nicely as a function of the grid size being processed. This is consistent with the IBM published results for this application.
  • The Total Application performance results are not typically reported in benchmark studies for this application. The IBM report specifically states that the execution times do not include I/O times and non-recurring allocation or initialization delays. Examining the total application performance reveals that the workload is no longer dominated by the the partial differential equation (PDE) solver, as IBM suggests, but is constrained by the I/O for grid initialization, reading in the traces, saving/restoring wave state data, and writing out the final image. Aggressive optimization of the PDE solver has little effect on the overall throughput of this application. It is clearly more important to optimize the I/O. The trend in seismic processing, as stated at the 2008 Society of Exploration Geophysicists (SEG) conference, is to run the reverse time migration iteratively on wide azimuth data. Thus, optimizing the I/O and application throughput is imperative to meet this trend. SSD and Flash technologies in conjunction with Sun's Lustre file system can reduce this I/O bottleneck and pave the path for the future in seismic processing.
  • Minimal tuning effort was applied to achieve the results presented. Sun's HPC software stack, which includes the Sun Studio compiler, was used to build the 70000 lines of C++ and Fortran source into the application executable. The only compiler option used was "-fast". No assembly level optimizations, like those performed by IBM to use SIMD registers (SSE registers), where performed in this benchmark. Similarly, no explicit cache blocking, loop unrolling, or memory bandwidth optimizations were conducted. The idea was to demonstrate the performance that a customer can expect from their existing applications without extensive, platform specific optimizations.

See Also

Disclosure Statement

Reverse Time Migration, Results as of 10/23/2009. For more info http://www.caam.rice.edu/tech_reports/2006/TR06-18.pdf

Wednesday Oct 14, 2009

Oracle Open World (OOW) BestPerf Index 14 October 2009

Here is a BestPerf blog index to a variety of benchmarks announced at Oracle Open World and others talked about at the conference.

Colors used:

Benchmark
Best Practices
Other

ORACLEOPENWORLD

CMT Servers

Oct 11, 2009 \* TPC-C World Record Sun - Oracle \*
Oct 13, 2009 Sun T5440 Oracle BI EE Sun T5440 World Record
Oct 13, 2009 SPECweb200 Sun T5440 World Record, Solaris Containers and Sun Storage F5100
Sep 01, 2009 String Searching - Sun T5240 & T5440 Outperform IBM Cell Broadband Engine
Aug 27, 2009 Sun T5240 Beats 4-Chip IBM Power 570 POWER6 System on SPECjbb2005
Aug 26, 2009 Sun T5220 Sets Single Chip World Record on SPECjbb2005
Aug 12, 2009 SPECmail2009 on Sun T5240 and Sun Java System Messaging Server 6.3
Jul 23, 2009 World Record Performance of Sun CMT Servers
Jul 22, 2009 Why does 1.6 beat 4.7?
Jul 21, 2009 Zeus ZXTM Traffic Manager World Record on Sun T5240
Jul 21, 2009 Sun T5440 World Record SAP-SD 4-Processor Two-tier SAP ERP 6.0 EP4 (Unicode)

SPARC64 Servers

Oct 13, 2009 SAP 2-tier SD Benchmark on Sun M9000/32 SPARC64 VII
Oct 13, 2009 Oracle PeopleSoft Payroll Sun M4000 and Sun Storage F5100 World Record Performance
Oct 12, 2009 Best Practices: M4000 Sun Storage F5100 is a good option for Peoplesoft Payroll
Oct 13, 2009 Oracle Hyperion Sun M5000 and Sun Storage 7410
Oct 13, 2009 SPECcpu2006 Results On MSeries Servers, New SPARC64 VII

X86 Servers

Oct 13, 2009 SAP 2-tier SD-Parallel on Sun Blade X6270 1-node, 2-node and 4-node
Aug 28, 2009 Sun X4270 World Record SAP-SD 2-Processor Two-tier SAP ERP 6.0 EP 4 (Unicode)
Oct 02, 2009 Sun X4270 VMware VMmark benchmark achieves excellent result
Sep 22, 2009 Sun X4270 Virtualized for Two-tier SAP ERP 6.0 EP4 (Unicode) Standard Sales and Distribution Benchmark

HPC Benchmarks

Oct 13, 2009 Halliburton ProMAX Oil & Gas Appl on Sun 6048/X6275 Cluster and Oracle Database
Oct 13, 2009 MCAE ABAQUS faster on Sun F5100 and Sun X4270 - Single Node World Record
Oct 12, 2009 MCAE ANSYS faster on Sun F5100 and Sun X4270
Oct 12, 2009 MCAE MCS/NASTRAN faster on Sun F5100 and Fire X4270
Oct 13, 2009 CP2K Life Sciences, Ab-initio Chem - Sun C48 with Sun Blade X6275 - QDR InfiniBand
Oct 09, 2009 X6275 Cluster Demonstrates Performance and Scalability on WRF 2.5km CONUS Dataset

Specific Storage Benchmarks

Oct 12, 2009 SPC-2 Sun Storage 6180 RAID 5 & RAID 6 Over 70% Better $/Performance than IBM
Oct 12, 2009 SPC-1 Sun Storage 6180 Over 70% Better $/Performance than IBM
Oct 12, 2009 1.6 Million 4K IOPS in 1RU on Sun Storage F5100 Flash Array

Additional CMT Server Benchmarks

Jul 21, 2009 1.6 GHz SPEC CPU2006 - Rate Benchmarks
Jul 21, 2009 Sun Blade T6320 World Record SPECjbb2005 performance
Jul 21, 2009 Sun T5440 SPECjbb2005 Beats IBM POWER6 Chip-to-Chip

Tuesday Oct 13, 2009

SAP 2-tier SD-Parallel on Sun Blade X6270 1-node, 2-node and 4-node

Significance of Results

  • Four Sun Blade X6270 (2 processors, 8 cores, 16 threads), running SAP ERP application Release 6.0 Enhancement Pack 4 (Unicode) with Oracle Database on top of Solaris 10 OS delivered the highest eight-processor result on the two-tier SAP SD-Parallel Standard Application Benchmark, as of Oct 12th, 2009.

  • Four Sun Blade X6270 servers with Intel Xeon X5570 processors achieved 1.9x performance improvement from two Sun Blade X6270 with the same processors.

  • Two Sun Blade X6270 (2 processors, 8 cores, 16 threads), running SAP ERP application Release 6.0 Enhancement Pack 4 (Unicode) with Oracle Database on top of Solaris 10 OS delivered the highest four-processor result on the two-tier SAP SD-Parallel Standard Application Benchmark, as of Oct 12th, 2009.

  • Two Sun Blade X6270 servers with Intel Xeon X5570 processors achieved 1.9x performance imporvement over a single 2-processor Sun Blade X6270 system.

  • A one node Sun Blade X6270 server with Intel Xeon X5570 processors running Oracle RAC delivers the same result as a Sun Fire X4270 server with Intel Xeon X5570 processors running Oracle with no performance difference between Oracle 10g and Oracle 10g RAC.

  • This benchmark highlights the near-linear scaling of Oracle 10g Real Application Cluster runs on Sun Microsystems hardware in a SAP environment.

  • In January 2009, a new version, the Two-tier SAP ERP 6.0 Enhancement Pack 4 (Unicode) Standard Sales and Distribution (SD) Benchmark, was released. This new release has higher cpu requirements and so yields from 25-50% fewer users compared to the previous Two-tier SAP ERP 6.0 (non-unicode) Standard Sales and Distribution (SD) Benchmark. 10-30% of this is due to the extra overhead from the processing of the larger character strings due to Unicode encoding. See this SAP Note for more details.

  • Unicode is a computing standard that allows for the representation and manipulation of text expressed in most of the world's writing systems. Before the Unicode requirement, this benchmark used ASCII characters meaning each was just 1 byte. The new version of the benchmark requires Unicode characters and the Application layer (where ~90% of the cycles in this benchmark are spent) uses a new encoding, UTF-16, which uses 2 bytes to encode most characters (including all ASCII characters) and 4 bytes for some others. This requires computers to do more computation and use more bandwidth and storage for most character strings. Refer to the above SAP Note for more details.

Performance Landscape

SAP SD-Parallel 2-Tier Performance Table (in decreasing performance order).

SAP ERP 6.0 Enhancement Pack 4 (Unicode) Results
(New version of the benchmark as of January 2009)

System OS
Database
Users SAP
ERP/ECC
Release
SAPS Date
Four Sun Blade X6270
2xIntel Xeon X5570 @2.93GHz
48 GB
Solaris 10
Oracle 10g Real Application Clusters
13,718 2009
6.0 EP4
(Unicode)
75,762 12-Oct-09
Two Sun Blade X6270
2xIntel Xeon X5570 @2.93GHz
48 GB
Solaris 10
Oracle 10g Real Application Clusters
7,220 2009
6.0 EP4
(Unicode)
39,420 12-Oct-09
One Sun Blade X6270
2xIntel Xeon X5570 @2.93GHz
48 GB
Solaris 10
Oracle 10g Real Application Clusters
3,800 2009
6.0 EP4
(Unicode)
20,750 12-Oct-09
Sun Fire X4270
2xIntel Xeon X5570 @2.93GHz
48 GB
Solaris 10
Oracle 10g
3,800 2009
6.0 EP4
(Unicode)
21,000 21-Aug-09

Complete benchmark results may be found at the SAP benchmark website http://www.sap.com/benchmark.

Results and Configuration Summary

Four Sun Blade X6270 Servers, each with two Intel Xeon X5570 2.93 GHz(2 processors, 8 cores, 16 threads)

    Number of SAP SD benchmark users:
    13,718
    Average dialog response time:
    0.86 seconds
    Throughput:

    Dialog steps/hour:
    4,545,729

    SAPS:
    75,762
    SAP Certification:
    2009041

Two Sun Blade X6270 Servers, each with two Intel Xeon X5570 2.93 GHz(2 processors, 8 cores, 16 threads)

    Number of SAP SD benchmark users:
    7,220
    Average dialog response time:
    0.99 seconds
    Throughput:

    Dialog steps/hour:
    2,365,000

    SAPS:
    39,420
    SAP Certification:
    2009040

One Sun Blade X6270 Servers, with two Intel Xeon X5570 2.93 GHz(2 processors, 8 cores, 16 threads)

    Number of SAP SD benchmark users:
    3,800
    Average dialog response time:
    0.99 seconds
    Throughput:

    Dialog steps/hour:
    1,245,000

    SAPS:
    20,750
    SAP Certification:
    2009039

Software:

    Oracle 10g Real Application Clusters
    Solaris 10 OS

Benchmark Description

The SAP Standard Application Sales and Distribution - Parallel (SD-Parallel) Benchmark is a two-tier ERP business test that is indicative of full business workloads of complete order processing and invoice processing, and demonstrates the ability to run both the application and database software on a single system. The SAP Standard Application SD Benchmark represents the critical tasks performed in real-world ERP business environments.
SD Versus SD-Parallel
The SD-Parallel Benchmark consists of the same transactions and user interaction steps as the SD Benchmark. This means that the SD-Parallel Benchmark runs the same business processes as the SD Benchmark. The difference between the benchmarks is the technical data distribution. An Additional Rule for Parallel and Distributed Databases
The additional rule is: Equally distribute the benchmark users across all database nodes for the used benchmark clients (round-robin-method). Following this rule, all database nodes work on data of all clients. This avoids unrealistic configurations such as having only one client per database node.
The SAP Benchmark Council agreed to give the parallel benchmark a different name so that the difference can be easily recognized by any interested parties - customers, prospects, and analysts. The naming convention is SD-Parallel for Sales & Distribution - Parallel.
SAP is one of the premier world-wide ERP application providers, and maintains a suite of benchmark tests to demonstrate the performance of competitive systems on the various SAP products.

Disclosure Statement

SAP SD benchmark based on SAP enhancement package 4 for SAP ERP 6.0 (Unicode) application benchmark as of Oct 12th, 2009: Four Sun Blade X6270 (each 2 processors, 8 cores, 16 threads) 13,718 SAP SD Users, 2x 2.93 GHz Intel Xeon x5570, each 48 GB memory, running two-tier SAP Sales and Distribution Parallel (SD-Parallel) standard SAP SD benchmark with Oracle 10g Real Application Clusters and Solaris 10, Cert# 2009041. Two Sun Blade X6270 (each 2 processors, 8 cores, 16 threads) 7,220 SAP SD Users, 2x 2.93 GHz Intel Xeon x5570, each 48 GB memory, running two-tier SAP Sales and Distribution Parallel (SD-Parallel) standard SAP SD benchmark with Oracle 10g Real Application Clusters and Solaris 10, Cert# 2009040. Sun Blade X6270 (2 processors, 8 cores, 16 threads) 3,800 SAP SD Users, 2x 2.93 GHz Intel Xeon x5570, 48 GB memory, running two-tier SAP Sales and Distribution Parallel (SD-Parallel) standard SAP SD benchmark with Oracle 10g Real Application Clusters and Solaris 10, Cert# 2009039. Sun Fire X4270 (2 processors, 8 cores, 16 threads) 3,800 SAP SD Users, 2x 2.93 GHz Intel Xeon x5570, 48 GB memory, running two-tier SAP Sales and Distribution (SD) standard SAP SD benchmark with Oracle 10g and Solaris 10, Cert# 2009033.

SAP, R/3, reg TM of SAP AG in Germany and other countries. More info www.sap.com/benchmark

Halliburton ProMAX Oil & Gas Application Fast on Sun 6048/X6275 Cluster

Significance of Results

The ProMAX family of seismic data processing tools is the most widely used Oil and Gas Industry seismic processing application. ProMAX is used for multiple applications, from field processing and quality control, to interpretive project-oriented reprocessing at oil companies and production processing at service companies. ProMAX is integrated with Halliburton's OpenWorks Geoscience Oracle Database to access prestack seismic data and populate the database with seismic images. This shows the powerful combination of scientific computing merged with commercial database technology.

A cluster of 48 Sun Blade X6275 server modules in a Sun Blade 6048 Modular System was configured with QDR Infiniband and a Lustre File System to demonstrate performance on ProMAX.

  • The 3D Prestack Kirchhoff Time Migration showed excellent scalability while utilizing the QDR Infiniband Lustre Filesystem.

    • 70808 Traces : 144x improvement going from 1 to 72 nodes
    • 283232 Traces : 98x improvement going from 1 to 96 nodes

  • The super linear scalability is attributed in part to data caching effects.

  • Improved the performance of the current production release of the ProMAX 3D Kirchhoff Time Migration by a factor up to 1.7x through recompilation of the source code using the Intel 11.1 compilers.

High Performance ProMAX allows Halliburton's GeoProbe interpretation application to perform migrations "on the fly" while pulling additional mapping data, well logs, and reservoir data from the OpenWorks Oracle Database.

  • Improves velocity modeling throughput for performing iterative Kirchhoff Migrations

  • Sun Grid Engine can be used to optimize the throughput of multiple migrations and maximize the return on investment of a Sun Blade 6048 Modular System.

Enabling hyperthreading and running 16 threads per node can benefit current and potential ProMAX users running on a Sun Blade X6275 configuration. In the tests run with the code rebuild, hyperthreading outperformed non-hyperthreading by as much as 27%.

Performance Landscape

Note: Results are all run with 16 Threads per Node with HyperThreading Enabled.


ProMAX 3D Prestack Kirchhoff Time Migration
SMP Threads and PVM Mode
Nodes Procs 70808 Traces 283232 Traces
Current Release
Execution Time
(sec)
Code Rebuild
Execution Time
(sec)
Current Release
Execution Time
(sec)
Code Rebuild
Execution Time
(sec)
96 192

18 13
72 144 3 2 22 16
48 96 4 3 38 23
24 48 11 7 76 48
16 32 23 14 117 72
12 24 37 23 165 100
8 16 62 35 258 150
4 8 129 78 514 343
1 2 486 288 2022 1278


ProMAX 3D Prestack Kirchhoff Time Migration
Scalability
Nodes Procs 70808 Traces 283232 Traces
Current Release
Speedup
1-Node
Code Rebuild
Speedup
1-Node
Current Release
Speedup
1-Node
Code Rebuild
Speedup
1-Node
96 192

112 98
72 144 162 144 92 80
48 96 121 96 53 55
24 48 44 41 26 26
16 32 21 20 17 18
12 24 13 12 12 13
8 16 8 8 8 8
4 8 4 4 4 4
1 2 1 1 1 1



ProMAX 3D Prestack Kirchhoff Time Migration
283232 Traces
Hyperthreading Performance Comparison
Nodes Procs Current Release Code Rebuild
8 Threads per
Node Perf
(sec)
16 Threads per
Node Perf
(sec)
8 Threads per
Node Perf
(sec)
16 Threads per
Node Perf
(sec)
24 48 95 76 59 48
16 32 144 117 91 72
12 24 196 165 122 100
8 16 310 258 190 150
4 8 629 514 390 343
1 2 2518 2022 1554 1278

Results and Configuration Summary

Hardware Configuration:
    Sun Blade 6048 Modular System with
      48 x Sun Blade X6275 (Vayu) Server Modules, each with
        4 x 2.93 GHz Intel Xeon QC X5570 processors
        12 x 4 GB memory at 1333 MHz
        2 x 24 GB Internal Flash
    QDR InfiniBand
    Lustre 1.8.0.1 File System

Software Configuration:

    OS: 64-bit SUSE Linux Enterprise Server SLES 10 SP 2
    PVM: Parallel Virtual Machine
    Resource Management: Sun Grid Engine
    Compiler: GNU C++ 4.1.2, Intel 11.1 Compilers
    Database: OpenWorks Database requires Oracle 10g Enterprise Edition
    Additional Libraries: pthreads 2.4, Java 1.6.0_01, BLAS, Stanford Exploration Project Libraries

Benchmark Description

This benchmark compares the current production release of ProMAX built with the GNU C++ and Fortran compilers to builds with the Intel Fortran and C++ compilers. Two different problem sizes were evaluated with the ProMAX 3D Prestack Kirchhoff Time Migration:
  • 70808 traces with 8 msec sample interval and trace length of 4992 msec
  • 283232 traces with 8 msec sample interval and trace length of 4992 msec

The ProMAX processing parameters used for this benchmark:
  • Input data set = Shots
    Minimum output inline = 65
    Maximum output inline = 85
    Inline output sampling interval = 1
    Minimum output xline = 1
    Maximum output xline = 200
    Xline output sampling interval = 1
    Antialias inline spacing = 15
    Antialias xline spacing = 15
    Stretch Mute Aperature Limit with Maximum Stretch = 15
    Image Gather Type = Full Offset Image Traces
    No Block Moveout
    Number of Alias Bands = 10
    3D Amplitude Phase Correction
    No compression
    Maximum Number of Cache Blocks = 500000

The compiler flags used for the various builds:
  • The Current Production Release Code was built with GNU Fortran and C++ flags.
    -O3 -m64 -march=x86-64 -mieee-fp -mfpmath=sse -msse2 -fforce-addr -fno-inline-functions

  • The application was rebuilt with the Intel Fortran and C++ flags.
    -xSSE4.2 -O3 -ipo -no-prec-div -static -m64 -ftz -fast-transcendentals -fp-speculation=fast

Key Points and Best Practices

Super linear scalability of the 70808 trace case for the larger node runs can be attributed to the fact that the dataset decomposition fits in cache which is shared by multiple threads per core.

See Also

Disclosure Statement

The following are trademarks or registered trademarks of Halliburton/Landmark Graphics: ProMAX, GeoProbe, OpenWorks.

Monday Oct 12, 2009

MCAE ABAQUS faster on Sun F5100 and Sun X4270 - Single Node World Record

The Sun Storage F5100 Flash Array can substantially improve performance over internal hard disk drives as shown by the I/O intensive ABAQUS MCAE application Standard benchmark tests on a Sun Fire X4270 server.

The I/O intensive ABAQUS "Standard" benchmarks test cases were run on a single Sun Fire X4270 server. Data is presented for runs at both 8 and 16 thread counts.

The ABAQUS "Standard" module is an MCAE application based on the finite element method (FEA) of analysis. This computer based numerical method inherently involves a substantial I/O component. The purpose was to evaluate the performance of the Sun Storage F5100 Flash Array relative to high performance 15K RPM internal striped HDDs.

  • The Sun Storage F5100 Flash Array outperformed the high performance 15K RPM SAS drives on the "S4b" test case by 14%.

  • The Sun Fire X4270 server coupled with a Sun Storage F5100 Flash Array established the world record performance on a single node for the four test cases S2A, S4B, S4D and S6.

Performance Landscape

ABAQUS "Standard" Benchmark Test S4B: Advantage of Sun Storage F5100

Results are total elapsed run times in seconds

Threads 4x15K RPM
72 GB SAS HDD
striped HW RAID0
Sun F5100
r/w buff 4096
striped
Sun F5100
Performance
Advantage
8 1504 1318 14%
16 1811 1649 10%

ABAQUS Standard Server Benchmark Subset: Single Node Record Performance

Results are total elapsed run times in seconds

Platform Cores S2a S4b S4d S6
X4270 w/F5100 8 302 1192 779 1237
HP BL460c G6 8 324 1309 843 1322
X4270 w/F5100 4 552 1970 1181 1706
HP BL460c G6 4 561 2062 1234 1812

Results and Configuration Summary

Hardware Configuration:
    Sun Fire X4270
      2 x 2.93 GHz QC Intel Xeon X5570 processors
      Hyperthreading enabled
      24 GB memory
      4 x 72 GB 15K RPM striped (HW RAID0) SAS disks
    Sun Storage F5100 Flash Array
      20 x 24 GB flash modules
      Intel controller

Software Configuration:

    O/S: 64-bit SUSE Linux Enterprise Server 10 SP 2
    Application: ABAQUS V6.9-1 Standard Module
    Benchmark: ABAQUS Standard Benchmark Test Suite

Benchmark Description

Abaqus/Standard Benchmark Problems

These problems provide an estimate of the performance that can be expected when running Abaqus/Standard or similar commercially available MCAE (FEA) codes like ANSYS and MSC/Nastran on different computers. The jobs are representative of those typically analyzed by Abaqus/Standard and other MCAE applications. These analyses include linear statics, nonlinear statics, and natural frequency extraction.

Please go here for a more complete description of the tests.

Key Points and Best Practices

  • The memory requirements for the test cases in the ABAQUS Standard benchmark test suite are rather substantial with some of the test cases requiring slightly over 20GB of memory. There are two memory limits one a minimum where out of core "memory" will be used when this limit is exceeded. This requires more time consuming cpu and another maximum memory limit that minimizes I/O operations. These memory limits are given in the ABAQUS output and can be established before making a full execution in a preliminary diagnostic mode run.
  • Based on the maximum physical memory on a platform the user can stipulate the maximum portion of this memory that can be allocated to the ABAQUS job. This is done in the "abaqus_v6.env" file that either resides in the subdirectory from where the job was launched or in the abaqus "site" subdirectory under the home installation directory.
  • Sometimes when running multiple cores on a single node, it is preferable from a performance standpoint to run in "smp" shared memory mode This is specified using the "THREADS" option on the "mpi_mode" line in the abaqus_v6.env file as opposed to the "MPI" option on this line. The test case considered here illustrates this point.
  • The test cases for the ABAQUS standard module all have a substantial I/O component where 15% to 25% of the total run times are associated with I/O activity (primarily scratch files). Performance will be enhanced by using the fastest available drives and striping together more than one of them or using a high performance disk storage system with high performance interconnects. On Linux OS's advantage can be taken of excess memory that can be used to cache and accelerate I/O.

See Also

Disclosure Statement

The following are trademarks or registered trademarks of Abaqus, Inc. or its subsidiaries in the United States and/or o ther countries: Abaqus, Abaqus/Standard, Abaqus/Explicit. All information on the ABAQUS website is Copyrighted 2004-2009 by Dassault Systemes. Results from http://www.simulia.com/support/v69/v69_performance.php as of October 12, 2009.

MCAE MCS/NASTRAN faster on Sun F5100 and Fire X4270

Significance of Results

The Sun Storage F5100 Flash Array can double performance over internal hard disk drives as shown by the I/O intensive MSC/Nastran MCAE application MDR3 benchmark tests on a Sun Fire X4270 server.

The MD Nastran MDR3 benchmarks were run on a single Sun Fire X4270 server. The I/O intensive test cases were run at different core levels from one up to the maximum of 8 available cores in SMP mode.

The MSC/Nastran MD 2008 R3 module is an MCAE application based on the finite element method (FEA) of analysis. This computer based numerical method inherently involves a substantial I/O component. The purpose was to evaluate the performance of the Sun Storage F5100 Flash Array relative to high performance 15K RPM internal stripped HDDs.

  • The Sun Storage F5100 Flash Array outperformed the high performance 15K RPM SAS drives on the "xx0cmd2" test case by 107% in the 8-core server configuration.

  • The Sun Storage F5100 Flash Array outperformed the high performance 15K RPM SAS drives on the "xl0tdf1"test case by 85% in the 8-core server configuration.

The MD Nastran MDR3 test suite was designed to include some very I/O intensive test cases albeit some are not very scalable. These cases are the called "xx0wmd0" and "xx0xst0". Both were run and results are presented using a single core server configuration.

  • The Sun Storage F5100 Flash Array outperformed the high performance 15K RPM SAS drives on the "xx0xst0"test case by 33% in the single-core server configuration.

  • The Sun Storage F5100 Flash Array outperformed the high performance 15K RPM SAS drives on the "xx0wmd0"test case by 20% in the single-core server configuration.

Performance Landscape

MD Nastran MDR3 Benchmark Tests

Results in seconds

Test Case DMP 4x15K RPM
72 GB SAS HDD
striped HW RAID0
Sun F5100
r/w buff 4096
striped
Sun F5100
Performance
Advantage
xx0cmd2 8 959 463 107%
xl0tdf1 8 1104 596 85%
xx0xst0 1 1307 980 33%
xx0wmd0 1 20250 16806 20%

Results and Configuration Summary

Hardware Configuration:
    Sun Fire X4270
      2 x 2.93 GHz QC Intel Xeon X5570 processors
      24 GB memory
      4 x 72 GB 15K RPM striped (HW RAID0) SAS disks
    Sun Storage F5100 Flash Array
      20 x 24 GB flash modules
      Intel controller

Software Configuration:

    O/S: 64-bit SUSE Linux Enterprise Server 10 SP 2
    Application: MSC/NASTRAN MD 2008 R3
    Benchmark: MDR3 Benchmark Test Suite
    HP MPI: 02.03.00.00 [7585] Linux x86-64

Benchmark Description

The benchmark tests are representative of typical MSC/Nastran applications including both SMP and DMP runs involving linear statics, nonlinear statics, and natural frequency extraction.

The MD (Multi Discipline) Nastran 2008 application performs both structural (stress) analysis and thermal analysis. These analyses may be either static or transient dynamic and can be linear or nonlinear as far as material behavior and/or deformations are concerned. The new release includes the MARC module for general purpose nonlinear analyses and the Dytran module that employs an explicit solver to analyze crash and high velocity impact conditions.

Please go here for a more complete description of the tests.

Key Points and Best Practices

  • Based on the maximum physical memory on a platform the user can stipulate the maximum portion of this memory that can be allocated to the Nastran job. This is done on the command line with the mem= option. On Linux based systems where the platform has a large amount of memory and where the model does not have large scratch I/O requirements the memory can be allocated to a tmpfs scratch space file system. On Solaris X64 systems advantage can be taken of ZFS for higher I/O performance.

  • The MD Nastran MDR3 test cases don't scale very well, a few not at all and the rest on up to 8 cores at best.

  • The test cases for the MSC/Nastran module all have a substantial I/O component where 15% to 25% of the total run times are associated with I/O activity (primarily scratch files). The required scratch file size ranges from less than 1 GB on up to about 140 GB. Performance will be enhanced by using the fastest available drives and striping together more than one of them or using a high performance disk storage system, further enhanced as indicated here by implementing the Lustre based I/O system. High performance interconnects such as InfiniBand for inter node cluster message passing as well as I/O transfer from the storage system can also enhance performance substantially.

See Also

Disclosure Statement

MSC.Software is a registered trademark of MSC. All information on the MSC.Software website is copyrighted. MD Nastran MDR3 results from http://www.mscsoftware.com and this report as of October 12, 2009.

About

BestPerf is the source of Oracle performance expertise. In this blog, Oracle's Strategic Applications Engineering group explores Oracle's performance results and shares best practices learned from working on Enterprise-wide Applications.

Index Pages
Search

Archives
« February 2016
SunMonTueWedThuFriSat
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
     
       
Today