Tuesday Feb 18, 2014

SPARC T5-2 Produces SPECjbb2013-MultiJVM World Record for 2-Chip Systems

From www.spec.org

Defects Identified in SPECjbb®2013

December 9, 2014 - SPEC has identified a defect in its SPECjbb®2013 benchmark suite. SPEC has suspended sales of the benchmark software and is no longer accepting new submissions of SPECjbb®2013 results for publication on SPEC's website. Current SPECjbb®2013 licensees will receive a free copy of the new version of the benchmark when it becomes available.

SPEC is advising SPECjbb®2013 licensees and users of the SPECjbb®2013 metrics that the recently discovered defect impacts the comparability of results. This defect can significantly impact the amount of work done during the measurement period, resulting in an inflated SPECjbb®2013 metric. SPEC recommends that users not utilize these results for system comparisons without a full understanding of the impact of these defects on each benchmark result.

Additional information is available here.

The SPECjbb2013 benchmark shows modern Java application performance. Oracle's SPARC T5-2 set a two-chip world record, which is 1.8x faster than the best two-chip x86-based server. Using Oracle Solaris and Oracle Java, Oracle delivered this two-chip world record result on the MultiJVM SPECjbb2013 metric.

  • The SPARC T5-2 server achieved 114,492 SPECjbb2013-MultiJVM max-jOPS and 43,963 SPECjbb2013-MultiJVM critical-jOPS on the SPECjbb2013 benchmark. This result is a two-chip world record.

  • The SPARC T5-2 server running SPECjbb2013 is 1.8x faster than the Cisco UCS C240 M3 server (2.7 GHz Intel Xeon E5-2697 v2) based on both the SPECjbb2013-MultiJVM max-jOPS and SPECjbb2013-MultiJVM critical-jOPS metrics.

  • The SPARC T5-2 server running SPECjbb2013 is 2x faster than the HP ProLiant ML350p Gen8 server (2.7 GHz Intel Xeon E5-2697 v2) based on SPECjbb2013-MultiJVM max-jOPS and 1.3x faster based on SPECjbb2013-MultiJVM critical-jOPS.

  • The new Oracle results were obtained using Oracle Solaris 11 along with Oracle Java SE 8 on the SPARC T5-2 server.

  • The SPARC T5-2 server running SPECjbb2013 on a per chip basis is 1.3x faster than the NEC Express5800/A040b server (2.8 GHz Intel Xeon E7-4890 v2) based on both the SPECjbb2013-MultiJVM max-jOPS and SPECjbb2013-MultiJVM critical-jOPS metrics.

  • There are no IBM POWER7 or POWER7+ based server results on the SPECjbb2013 benchmark. IBM has published IBM POWER7+ based servers on the SPECjbb2005 which was retired by SPEC in 2013.

Performance Landscape

Results of SPECjbb2013 from www.spec.org as of March 6, 2014. These are the leading 2-chip SPECjbb2013 MultiJVM results.

SPECjbb2013 - 2-Chip MultiJVM Results
System Processor SPECjbb2013-MultiJVM JDK
max-jOPS critical-jOPS
SPARC T5-2 2xSPARC T5, 3.6 GHz 114,492 43,963 Oracle Java SE 8
Cisco UCS C240 M3 2xIntel E5-2697 v2, 2.7 GHz 63,079 23,797 Oracle Java SE 7u45
HP ProLiant ML350p Gen8 2xIntel E5-2697 v2, 2.7 GHz 62,393 24,310 Oracle Java SE 7u45
IBM System x3650 M4 BD 2xIntel E5-2695 v2, 2.4 GHz 59,124 22,275 IBM SDK V7 SR6 (*)
HP ProLiant ML350p Gen8 2xIntel E5-2697 v2, 2.7 GHz 57,594 32,103 Oracle Java SE 7u40
HP ProLiant BL460c Gen8 2xIntel E5-2697 v2, 2.7 GHz 56,367 30,078 Oracle Java SE 7u40
Sun Server X4-2, DDR3-1600 2xIntel E5-2697 v2, 2.7 GHz 52,664 20,553 Oracle Java SE 7u40
HP ProLiant DL360e Gen8 2xIntel E5-2470 v2, 2.4 GHz 48,772 17,915 Oracle Java SE 7u40

* IBM SDK V7 SR6 – IBM SDK, Java Technology Edition, Version 7, Service Refresh 6

The following table compares the SPARC T5 processor to the Intel E7 v2 processor.

SPECjbb2013 - Results Using JDK 8
Per Chip Comparison
System SPECjbb2013-MultiJVM SPECjbb2013-MultiJVM/Chip JDK
max-jOPS critical-jOPS max-jOPS critical-jOPS
SPARC T5-2
2xSPARC T5, 3.6 GHz
114,492 43,963 57,246 21,981 Oracle Java SE 8
NEC Express5800/A040b
4xIntel E7-4890 v2, 2.8 GHz
177,753 65,529 44,438 16,382 Oracle Java SE 8

SPARC per Chip Advantage 1.29x 1.34x

Configuration Summary

System Under Test:

SPARC T5-2 server
2 x SPARC T5, 3.60 GHz
512 GB memory (32 x 16 GB dimms)
Oracle Solaris 11.1
Oracle Java SE 8

Benchmark Description

The SPECjbb2013 benchmark has been developed from the ground up to measure performance based on the latest Java application features. It is relevant to all audiences who are interested in Java server performance, including JVM vendors, hardware developers, Java application developers, researchers and members of the academic community.

From SPEC's press release, "SPECjbb2013 replaces SPECjbb2005. The new benchmark has been developed from the ground up to measure performance based on the latest Java application features. It is expected to be used widely by all those interested in Java server performance, including JVM vendors, hardware developers, Java application developers, researchers and members of the academic community."

SPECjbb2013 features include:

  • A usage model based on a world-wide supermarket company with an IT infrastructure that handles a mix of point-of-sale requests, online purchases and data-mining operations.
  • Both a pure throughput metric and a metric that measures critical throughput under service-level agreements (SLAs) specifying response times ranging from 10ms to 500ms.
  • Support for multiple run configurations, enabling users to analyze and overcome bottlenecks at multiple layers of the system stack, including hardware, OS, JVM and application layers.
  • Exercising new Java 7 features and other important performance elements, including the latest data formats (XML), communication using compression, and messaging with security.
  • Support for virtualization and cloud environments.

See Also

Disclosure Statement

SPEC and the benchmark name SPECjbb are registered trademarks of Standard Performance Evaluation Corporation (SPEC). Results as of 3/6/2014, see http://www.spec.org for more information.  SPARC T5-2 114,492 SPECjbb2013-MultiJVM max-jOPS, 43,963 SPECjbb2013-MultiJVM critical-jOPS; NEC Express5800/A040b 177,753 SPECjbb2013-MultiJVM max-jOPS, 65,529 SPECjbb2013-MultiJVM critical-jOPS; Cisco UCS c240 M3 63,079 SPECjbb2013-MultiJVM max-jOPS, 23,797 SPECjbb2013-MultiJVM critical-jOPS; HP ProLiant ML350p Gen8 62,393 SPECjbb2013-MultiJVM max-jOPS, 24,310 SPECjbb2013-MultiJVM critical-jOPS; IBM System X3650 M4 BD 59,124 SPECjbb2013-MultiJVM max-jOPS, 22,275 SPECjbb2013-MultiJVM critical-jOPS; HP ProLiant ML350p Gen8 57,594 SPECjbb2013-MultiJVM max-jOPS, 32,103 SPECjbb2013-MultiJVM critical-jOPS; HP ProLiant BL460c Gen8 56,367 SPECjbb2013-MultiJVM max-jOPS, 30,078 SPECjbb2013-MultiJVM critical-jOPS; Sun Server X4-2 52,664 SPECjbb2013-MultiJVM max-jOPS, 20,553 SPECjbb2013-MultiJVM critical-jOPS; HP ProLiant DL360e Gen8 48,772 SPECjbb2013-MultiJVM max-jOPS, 17,915 SPECjbb2013-MultiJVM critical-jOPS.

Friday Feb 14, 2014

SPARC M6-32 Delivers Oracle E-Business and PeopleSoft World Record Benchmarks, Linear Data Warehouse Scaling in a Virtualized Configuration

This result demonstrates how the combination of Oracle virtualization technologies for SPARC and Oracle's SPARC M6-32 server allow the deployment and concurrent high performance execution of multiple Oracle applications and databases sized for the Enterprise.

  • In an 8-chip Dynamic Domain (also known as PDom), the SPARC M6-32 server set a World Record E-Business 12.1.3 X-Large world record with 14,660 online users running five simultaneous E-Business modules.

  • In a second 8-chip Dynamic Domain, the SPARC M6-32 server set a World Record PeopleSoft HCM 9.1 HR Self-Service online supporting 35,000 users while simultaneously running a batch workload in 29.17 minutes. This was done with a database of 600,480 employees. Two other separate tests were run, one supporting 40,000 online users only and another a batch-only workload that was run in 18.27 min.

  • In a third Dynamic Domain with 16-chips on the SPARC M6-32 server, a data warehouse test was run that showed near-linear scaling.

  • On the SPARC M6-32 server, several critical applications instances were virtualized: an Oracle E-Business application and database, an Oracle's PeopleSoft application and database, and a Decision Support database instance using Oracle Database 12c.

  • In this Enterprise Virtualization benchmark a SPARC M6-32 server utilized all levels of Oracle Virtualization features available for SPARC servers. The 32-chip SPARC M6 based server was divided in three separate Dynamic Domains (also known as PDoms), available only on the SPARC Enterprise M-Series systems, which are completely electrically isolated and independent hardware partitions. Each PDom was subsequently split into multiple hypervisor-based Oracle VM for SPARC partitions (also known as LDoms), each one running its own Oracle Solaris kernel and managing its own CPUs and I/O resources. The hardware resources allocated to each Oracle VM for SPARC partition were then organized in various Oracle Solaris Zones, to further refine application tier isolation and resources management. The three PDoms were dedicated to the enterprise applications as follows:

    • Oracle E-Business PDom: Oracle E-Business 12.1.3 Suite World Record Extra-Large benchmark, exercising five Online Modules: Customer Service, Human Resources Self Service, iProcurement, Order Management and Financial, with 14,660 users and an average user response time under 2 seconds.

    • PeopleSoft PDom: PeopleSoft Human Capital Management (HCM) 9.1 FP2 World Record Benchmark, using PeopleTools 8.52 and an Oracle Database 11g Release 2, with 35,000 users, at an average user Search Time of 1.46 seconds and Save Time of 0.93 seconds. An online run with 40,000 users, had an average user Search Time of 2.17 seconds and Save Time of 1.39 seconds, and a Payroll batch run completed in 29.17 minutes elapsed time for more than 500,000 employees.

    • Decision Support PDom: An Oracle Database 12c instance executing a Decision Support workload on about 30 billion rows of data and achieving linear scalability, i.e. on the 16 chips comprising the PDom, the workload ran 16x faster than on a single chip. Specifically, the 16-chip PDom processed about 320M rows/sec whereas a single chip could process about 20M rows/sec.

  • The SPARC M6-32 server is ideally suited for large-memory utilization. In this virtualized environment, three critical applications made use of 16 TB of physical memory. Each of the Oracle VM Server for SPARC environments utilized from 4 to 8 TB of memory, more than the limits of other virtualization solutions.

  • SPARC M6-32 Server Virtualization Layout Highlights

    • The Oracle E-Business application instances were run in a dedicated Dynamic Domain consisting of 8 SPARC M6 processors and 4 TB of memory. The PDom was split into four symmetric Oracle VM Server for SPARC (LDoms) environments of 2 chips and 1 TB of memory each, two dedicated to the Application Server tier and the other two to the Database Server tier. Each Logical Domain was subsequently divided into two Oracle Solaris Zones, for a total of eight, one for each E-Business Application server and one for each Oracle Database 11g instance.

    • The PeopleSoft application was run in a dedicated Dynamic Domain (PDom) consisting of 8 SPARC M6 processors and 4 TB of memory. The PDom was split into two Oracle VM Server for SPARC (LDoms) environments one of 6 chips and 3 TB of memory, reserved for the Web and Application Server tiers, and a second one of 2 chips and 1 TB of memory, reserved for the Database tier. Two PeopleSoft Application Servers, a Web Server instance, and a single Oracle Database 11g instance were each executed in their respective and exclusive Oracle Solaris Zone.

    • The Oracle Database 12c Decision Support workload was run in a Dynamic Domain consisting of 16 SPARC M6 processors and 8 TB of memory.

  • All the Oracle Applications and Database instances were running at high level of performance and concurrently in a virtualized environment. Running three Enterprise level application environments on a single SPARC M6-32 server offers centralized administration, simplified physical layout, high availability and security features (as each PDom and LDom runs its own Oracle Solaris operating system copy physically and logically isolated from the other environments), enabling the coexistence of multiple versions Oracle Solaris and application software on a single physical server.

  • Dynamic Domains and Oracle VM Server for SPARC guests were configured with independent direct I/O domains, allowing for fast and isolated I/O paths, providing secure and high performance I/O access.

Performance Landscape

Oracle E-Business Test using Oracle Database 11g
SPARC M6-32 PDom, 8 SPARC M6 Processors, 4 TB Memory
Total Online Users Weighted Average
Response Time (sec)
90th Percentile
Response Time (s)
14,660 0.81 0.88
Multiple Online Modules X-Large Configuration (HR Self-Service, Order Management, iProcurement, Customer Service, Financial)

PeopleSoft HR Self-Service Online Plus Payroll Batch using Oracle Database 11g
SPARC M6-32 PDom, 8 SPARC M6 Processors, 4 TB Memory
HR Self-Service Payroll Batch
Elapsed (min)
Online Users Average User
Search / Save
Time (sec)
Transactions
per Second
35,000 1.46 / 0.93 116 29.17

HR Self-Service Only Payroll Batch Only
Elapsed (min)
40,000 2.17 / 1.39 132 18.27

Oracle Database 12c Decision Support Query Test
SPARC M6-32 PDom, 16 SPARC M6 Processors, 8 TB Memory
Parallelism
Chips Used
Rows Processing Rate
(rows/s)
Scaling Normalized to 1 Chip
16 319,981,734 15.9
8 162,545,303 8.1
4 80,943,271 4.0
2 40,458,329 2.0
1 20,086,829 1.0

Configuration Summary

System Under Test:

SPARC M6-32 server with
32 x SPARC M6 processors (3.6 GHz)
16 TB memory

Storage Configuration:

6 x Sun Storage 2540-M2 each with
8 x Expansion Trays (each tray equipped with 12 x 300 GB SAS drives)
7 x Sun Server X3-2L each with
2 x Intel Xeon E5-2609 2.4 GHz Processors
16 GB Memory
4 x Sun Flash Accelerator F40 PCIe 400 GB cards
Oracle Solaris 11.1 (COMSTAR)
1 x Sun Server X3-2L with
2 x Intel Xeon E5-2609 2.4 GHz Processors
16 GB Memory
12 x 3 TB SAS disks
Oracle Solaris 11.1 (COMSTAR)

Software Configuration:

Oracle Solaris 11.1 (11.1.10.5.0), Oracle E-Business
Oracle Solaris 11.1 (11.1.10.5.0), PeopleSoft
Oracle Solaris 11.1 (11.1.9.5.0), Decision Support
Oracle Database 11g Release 2, Oracle E-Business and PeopleSoft
Oracle Database 12c Release 1, Decision Support
Oracle E-Business Suite 12.1.3
PeopleSoft Human Capital Management 9.1 FP2
PeopleSoft PeopleTools 8.52.03
Oracle Java SE 6u32
Oracle Tuxedo, Version 10.3.0.0, 64-bit, Patch Level 043
Oracle WebLogic Server 11g (10.3.4)

Oracle Dynamic Domains (PDoms) resources:


Oracle E-Business PeopleSoft Oracle DSS
Processors 8 8 16
Memory 4 TB 4 TB 8 TB
Oracle Solaris 11.1 (11.1.10.5.0) 11.1 (11.1.10.5.0) 11.1 (11.1.9.5.0)
Oracle Database 11g 11g 12c
Oracle VM for SPARC /
Oracle Solaris Zones
4 LDom / 8 Zones 2 LDom / 4 Zones None
Storage 7 x Sun Server X3-2L 1 x Sun Server X3-2L
(12 x 3 TB SAS )
2 x Sun Storage 2540-M2 / 2501 pairs
4 x Sun Storage 2540-M2/2501 pairs

Benchmark Description

This benchmark consists of three different applications running concurrently. It shows that large, enterprise workloads can be run on a single system and without performance impact between application environments.

The three workloads are:

  • Oracle E-Business Suite Online

    • This test simulates thousands of online users executing transactions typical of an internal Enterprise Resource Processing, including 5 application modules: Customer Service, Human Resources Self Service, Procurement, Order Management and Financial.

    • Each database tier uses a database instance of about 600 GB in size, and supporting thousands of application users, accessing hundreds of objects (tables, indexes, SQL stored procedures, etc.).

    • The application tier includes multiple web and application server instances, specifically Apache Web Server, Oracle Application Server 10g and Oracle Java SE 6u32.

  • PeopleSoft Human Capital Management

    • This test simulates thousands of online employees, managers and Human Resource administrators executing transactions typical of a Human Resources Self Service application for the Enterprise. Typical transactions are: viewing paychecks, promoting and hiring employees, updating employee profiles, etc.

    • The database tier uses a database instance of about 500 GB in size, containing information for 500,480 employees.

    • The application tier for this test includes web and application server instances, specifically Oracle WebLogic Server 11g, PeopleSoft Human Capital Management 9.1 and Oracle Java SE 6u32.

  • Decision Support Workload using the Oracle Database.

    • The query processes 30 billion rows stored in the Oracle Database, making heavy use of Oracle parallel query processing features. It performs multiple aggregations and summaries by reading and processing all the rows of the database.

Key Points and Best Practices

Oracle E-Business Environment

The Oracle E-Business Suite setup consisted 4 Oracle E-Business environments running 5 online Oracle E-Business modules simultaneously.

The Oracle E-Business environments were deployed on 4 Oracle VM for SPARC, respectively 2 for the Application tier and 2 for the Database tier. Each LDom included 2 SPARC M6 processor chips. The Application LDom was further split into 2 Oracle Solaris Zones, each one containing one Oracle E-Business Application instance. Similarly, on the Database tier, each LDom was further divided into 2 Oracle Solaris Zones, each containing an Oracle Database instance. Applications on the same LDom shared a 10 GbE network link to connect to the Database tier LDom. Each Application in a Zone was connected to its own dedicated Database Zone. The communication between the two Zones was implemented via Oracle Solaris 11 virtual network, which provides high performance, low latency transfers at memory speed using large frames (9000 bytes vs typical 1500 bytes frames).

The Oracle E-Business setup made use of the Oracle Database Shared Server feature in order to limit memory utilization, as well as the number of database Server processes. The Oracle Database configuration and optimization was substantially out-of-the-box, except for proper sizing the Oracle Database memory areas (System Global Area and Program Global Area).

In the Oracle E-Business Application LDom handling Customer Service and HR Self Service modules, 28 Forms servers and 8 OC4J application servers were hosted in the two separate Oracle Solaris Zones, for a total of 56 forms servers and 16 applications servers.

All the Oracle Database server processes and the listener processes were executed in the Oracle Solaris FX scheduler class.

PeopleSoft Environment

The PeopleSoft Application Oracle VM for SPARC had one Oracle Solaris Zone of 12 cores containing the web tier and two Oracle Solaris Zones of 57 cores total containing the Application tier. The Database tier was contained in an Oracle VM for SPARC consisting of one Oracle Solaris Zone of 24 cores. One core, in the Application Oracle VM, was dedicated to network and disk interrupt handling.

All database data files, recovery files and Oracle Clusterware files for the PeopleSoft test were created with the Oracle Automatic Storage Management (Oracle ASM) volume manager for the added benefit of the ease of management provided by Oracle ASM integrated storage management solution.

In the application tier, 5 PeopleSoft domains with 350 application servers (70 per each domain) were hosted in the two separate Oracle Solaris Zones for a total of 10 domains with 700 application server processes.

All PeopleSoft Application processes and Web Server JVM instances were executed in the Oracle Solaris FX scheduler class.

Oracle Decision Support Environment

The decision support workload showed how the combination of a large memory (8 TB) and a large number of processors (16 chips comprising 1536 virtual CPUs) together with Oracle parallel query facility can linearly increase the performance of certain decision support queries as the number of CPUs increase.

The large memory was used to cache the entire 30 billion row Oracle table in memory. There are a number of ways to accomplish this. The method deployed in this test was to allocate sufficient memory for Oracle's "keep cache" and direct the table to the "keep cache."

To demonstrate scalability, it was necessary to ensure that the number of Oracle parallel servers was always equal to the number of available virtual CPUs. This was accomplished by the combination of providing a degree of parallelism hint to the query and setting both parallel_max_servers and parallel_min_servers to the number of virtual CPUs.

The number of virtual CPUs for each stage of the scalability test was adjusted using the psradm command available in Oracle Solaris.

See Also

Disclosure Statement

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. PeopleSoft results as of 02/14/2014. Other results as of 09/22/2013.

Oracle E-Business Suite R12 extra-large multiple-online module benchmark, SPARC M6-32, SPARC M6, 3.6 GHz, 8 chips, 96 cores, 768 threads, 4 TB memory, 14,660 online users, average response time 0.81 sec, 90th percentile response time 0.88 sec, Oracle Solaris 11.1, Oracle Solaris Zones, Oracle VM for SPARC, Oracle E-Business Suite 12.1.3, Oracle Database 11g Release 2, Results as of 9/22/2013.

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 2014 »
SunMonTueWedThuFriSat
      
1
2
3
4
5
6
7
8
9
10
11
12
13
15
16
17
19
20
21
22
23
24
25
26
27
28
 
       
Today