Wednesday Sep 25, 2013

SPARC T5-8 Delivers World Record Oracle OLAP Perf Version 3 Benchmark Result on Oracle Database 12c

Oracle's SPARC T5-8 server delivered world record query performance for systems running Oracle Database 12c for the Oracle OLAP Perf Version 3 benchmark.

  • The query throughput on the SPARC T5-8 server is 1.7x higher than that of an 8-chip Intel Xeon E7-8870 server. Both systems had sub-second average response times.

  • The SPARC T5-8 server with the Oracle Database demonstrated the ability to support at least 700 concurrent users querying OLAP cubes (with no think time), processing 2.33 million analytic queries per hour with an average response time of less than 1 second per query. This performance was enabled by keeping the entire cube in-memory utilizing the 4 TB of memory on the SPARC T5-8 server.

  • Assuming a 60 second think time between query requests, the SPARC T5-8 server can support approximately 39,450 concurrent users with the same sub-second response time.

  • The workload uses a set of realistic Business Intelligence (BI) queries that run against an OLAP cube based on a 4 billion row fact table of sales data. The 4 billion rows are partitioned by month spanning 10 years.

  • The combination of the Oracle Database 12cwith the Oracle OLAP option running on a SPARC T5-8 server supports live data updates occurring concurrently with minimally impacted user query executions.

Performance Landscape

Oracle OLAP Perf Version 3 Benchmark
Oracle cube base on 4 billion fact table rows
10 years of data partitioned by month
System Queries/
hour
Users Average Response
Time (sec)
0 sec think time 60 sec think time
SPARC T5-8 2,329,000 700 39,450 <1 sec
8-chip Intel Xeon E7-8870 1,354,000 120 22,675 <1 sec

Configuration Summary

SPARC T5-8:

1 x SPARC T5-8 server with
8 x SPARC T5 processors, 3.6 GHz
4 TB memory
Data Storage and Redo Storage
Flash Storage
Oracle Solaris 11.1 (11.1.8.2.0)
Oracle Database 12c Release 1 (12.1.0.1) with Oracle OLAP option

Sun Server X2-8:

1 x Sun Server X2-8 with
8 x Intel Xeon E7-8870 processors, 2.4 GHz
1 TB memory
Data Storage and Redo Storage
Flash Storage
Oracle Solaris 10 10/12
Oracle Database 12c Release 1 (12.1.0.1) with Oracle OLAP option

Benchmark Description

The Oracle OLAP Perf Version 3 benchmark is a workload designed to demonstrate and stress the ability of the OLAP Option to deliver fast query, near real-time updates and rich calculations using a multi-dimensional model in the context of the Oracle data warehousing.

The bulk of the benchmark entails running a number of concurrent users, each issuing typical multidimensional queries against an Oracle cube. The cube has four dimensions: time, product, customer, and channel. Each query user issues approximately 150 different queries. One query chain may ask for total sales in a particular region (e.g South America) for a particular time period (e.g. Q4 of 2010) followed by additional queries which drill down into sales for individual countries (e.g. Chile, Peru, etc.) with further queries drilling down into individual stores, etc. Another query chain may ask for yearly comparisons of total sales for some product category (e.g. major household appliances) and then issue further queries drilling down into particular products (e.g. refrigerators, stoves. etc.), particular regions, particular customers, etc.

While the core of every OLAP Perf benchmark is real world query performance, the benchmark itself offers numerous execution options such as varying data set sizes, number of users, numbers of queries for any given user and cube update frequency. Version 3 of the benchmark is executed with a much larger number of query streams than previous versions and used a cube designed for near real-time updates. The results produced by version 3 of the benchmark are not directly comparable to results produced by previous versions of the benchmark.

The near real-time update capability is implemented along the following lines. A large Oracle cube, H, is built from a 4 billion row star schema, containing data up until the end of last business day. A second small cube, D, is then created which will contain all of today's new data coming in from outside the world. It will be updated every L minutes with the data coming in within the last L minutes. A third cube, R, joins cubes H and D for reporting purposes much like a view might join data from two tables. Calculations are installed into cube R. The use of a reporting cube which draws data from different storage cubes is a common practice.

Query users are never locked out of query operations while new data is added to the update cube. The point of the demonstration is to show that an Oracle OLAP system can be designed which results in data being no more than L minutes out of date, where L may be as low as just a few minutes. This is what is meant by near real-time analytics.

Key Points and Best Practices

  • Building and querying cubes with the Oracle OLAP option requires a large temporary tablespace. Normally temporary tablespaces would reside on disk storage. However, because the SPARC T5-8 server used in this benchmark had 4 TB of main memory, it was possible to use main memory for the OLAP temporary tablespace. This was accomplished by using a temporary, memory-based file system (TMPFS) for the temporary tablespace datafiles.

  • Since typical business intelligence users are often likely to issue similar queries, either with the same or different constants in the where clauses, setting the init.ora parameter "cursor_sharing" to "force" provides for additional query throughput and a larger number of potential users.

  • Assuming the normal Oracle Database initialization parameters (e.g. SGA, PGA, processes etc.) are appropriately set, out of the box performance for the Oracle OLAP workload should be close to what is reported here. Additional performance resulted from using memory for the OLAP temporary tablespace setting "cursor_sharing" to force.

  • Oracle OLAP Cube update performance was optimized by running update processes in the FX class with a priority greater than 0.

  • The maximum lag time between updates to the source fact table and data availability to query users (what was referred to as L in the benchmark description) was less than 3 minutes for the benchmark environment on the SPARC T5-8 server.

See Also

Disclosure Statement

Copyright 2013, 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 09/22/2013.

SPARC T5 Encryption Performance Tops Intel E5-2600 v2 Processor

The cryptography benchmark suite was developed by Oracle to measure security performance on important AES security modes. Oracle's SPARC T5 processor with it security software in silicon is faster than x86 servers that have the AES-NI instructions. In this test, the performance of on-processor encryption operations is measured (32 KB encryptions). Multiple threads are used to measure each processors maximum throughput. The SPARC T5-8 shows dramatically faster encryption.

  • A SPARC T5 processor running Oracle Solaris 11.1 is 2.7 times faster executing AES-CFB 256-bit key encryption (in cache) than the Intel E5-2697 v2 processor (with AES-NI) running Oracle Linux 6.3. AES-CFB encryption is used by Oracle Database for Transparent Data Encryption (TDE) which provides security for database storage.

  • On the AES-CFB 128-bit key encryption, the SPARC T5 processor is 2.5 times faster than the Intel E5-2697 v2 processor (with AES-NI) running Oracle Linux 6.3 for in-cache encryption. AES-CFB mode is used by Oracle Database for Transparent Data Encryption (TDE) which provides security for database storage.

  • The IBM POWER7+ has three hardware security units for 8-core processors, but IBM has not publicly shown any measured performance results on AES-CFB or other encryption modes.

Performance Landscape

Presented below are results for running encryption using the AES cipher with the CFB, CBC, CCM and GCM modes for key sizes of 128, 192 and 256. Decryption performance was similar and is not presented. Results are presented as MB/sec (10**6).

Encryption Performance – AES-CFB

Performance is presented for in-cache AES-CFB128 mode encryption. Multiple key sizes of 256-bit, 192-bit and 128-bit are presented. The encryption was performance on 32 KB of pseudo-random data (same data for each run).

AES-CFB
Microbenchmark Performance (MB/sec)
Processor GHz Chips Performance Software Environment
AES-256-CFB
SPARC T5 3.60 2 54,396 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2697 v2 2.70 2 19,960 Oracle Linux 6.3, IPP/AES-NI
Intel E5-2690 2.90 2 12,823 Oracle Linux 6.3, IPP/AES-NI
AES-192-CFB
SPARC T5 3.60 2 61,000 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2697 v2 2.70 2 23,217 Oracle Linux 6.3, IPP/AES-NI
Intel E5-2690 2.90 2 14,928 Oracle Linux 6.3, IPP/AES-NI
AES-128-CFB
SPARC T5 3.60 2 68,695 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2697 v2 2.70 2 27,740 Oracle Linux 6.3, IPP/AES-NI
Intel E5-2690 2.90 2 17,824 Oracle Linux 6.3, IPP/AES-NI

Encryption Performance – AES-GCM

Performance is presented for in-cache AES-GCM mode encryption with authentication. Multiple key sizes of 256-bit, 192-bit and 128-bit are presented. The encryption/authentication was performance on 32 KB of pseudo-random data (same data for each run).

AES-GCM
Microbenchmark Performance (MB/sec)
Processor GHz Chips Performance Software Environment
AES-256-GCM
SPARC T5 3.60 2 34,101 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2697 v2 2.70 2 15,338 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2690 2.90 2 13,520 Oracle Linux 6.3, IPP/AES-NI
AES-192-GCM
SPARC T5 3.60 2 36,852 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2697 v2 2.70 2 15,768 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2690 2.90 2 14,159 Oracle Linux 6.3, IPP/AES-NI
AES-128-GCM
SPARC T5 3.60 2 39,003 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2697 v2 2.70 2 16,405 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2690 2.90 2 14,877 Oracle Linux 6.3, IPP/AES-NI

Encryption Performance – AES-CCM

Performance is presented for in-cache AES-CCM mode encryption with authentication. Multiple key sizes of 256-bit, 192-bit and 128-bit are presented. The encryption/authentication was performance on 32 KB of pseudo-random data (same data for each run).

AES-CCM
Microbenchmark Performance (MB/sec)
Processor GHz Chips Performance Software Environment
AES-256-CCM
SPARC T5 3.60 2 29,431 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2697 v2 2.70 2 19,447 Oracle Linux 6.3, IPP/AES-NI
Intel E5-2690 2.90 2 12,493 Oracle Linux 6.3, IPP/AES-NI
AES-192-CCM
SPARC T5 3.60 2 33,715 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2697 v2 2.70 2 22,634 Oracle Linux 6.3, IPP/AES-NI
Intel E5-2690 2.90 2 14,507 Oracle Linux 6.3, IPP/AES-NI
AES-128-CCM
SPARC T5 3.60 2 39,188 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2697 v2 2.70 2 26,951 Oracle Linux 6.3, IPP/AES-NI
Intel E5-2690 2.90 2 17,256 Oracle Linux 6.3, IPP/AES-NI

Encryption Performance – AES-CBC

Performance is presented for in-cache AES-CBC mode encryption. Multiple key sizes of 256-bit, 192-bit and 128-bit are presented. The encryption was performance on 32 KB of pseudo-random data (same data for each run).

AES-CBC
Microbenchmark Performance (MB/sec)
Processor GHz Chips Performance Software Environment
AES-256-CBC
SPARC T5 3.60 2 56,933 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2697 v2 2.70 2 19,962 Oracle Linux 6.3, IPP/AES-NI
Intel E5-2690 2.90 2 12,822 Oracle Linux 6.3, IPP/AES-NI
AES-192-CBC
SPARC T5 3.60 2 63,767 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2697 v2 2.70 2 23,224 Oracle Linux 6.3, IPP/AES-NI
Intel E5-2690 2.90 2 14,915 Oracle Linux 6.3, IPP/AES-NI
AES-128-CBC
SPARC T5 3.60 2 72,508 Oracle Solaris 11.1, libsoftcrypto + libumem
Intel E5-2697 v2 2.70 2 27,733 Oracle Linux 6.3, IPP/AES-NI
Intel E5-2690 2.90 2 17,823 Oracle Linux 6.3, IPP/AES-NI

Configuration Summary

SPARC T5-2 server
2 x SPARC T5 processor, 3.6 GHz
512 GB memory
Oracle Solaris 11.1 SRU 4.2

Sun Server X4-2L server
2 x E5-2697 v2 processors, 2.70 GHz
256 GB memory
Oracle Linux 6.3

Sun Server X3-2 server
2 x E5-2690 processors, 2.90 GHz
128 GB memory
Oracle Linux 6.3

Benchmark Description

The benchmark measures cryptographic capabilities in terms of general low-level encryption, in-cache (32 KB encryptions) and on-chip using various ciphers, including AES-128-CFB, AES-192-CFB, AES-256-CFB, AES-128-CBC, AES-192-CBC, AES-256-CBC, AES-128-CCM, AES-192-CCM, AES-256-CCM, AES-128-GCM, AES-192-GCM and AES-256-GCM.

The benchmark results were obtained using tests created by Oracle which use various application interfaces to perform the various ciphers. They were run using optimized libraries for each platform to obtain the best possible performance.

See Also

Disclosure Statement

Copyright 2013, 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 9/23/2013.

Sun Server X4-2 Delivers Single App Server, 2-Chip x86 World Record SPECjEnterprise2010

Oracle's Sun Server X4-2 and Sun Server X4-2L servers, using the Intel Xeon E5-2697 v2 processor, produced a world record x86 two-chip single application server SPECjEnterprise2010 benchmark result of 11,259.88 SPECjEnterprise2010 EjOPS. The Sun Server X4-2 ran the application tier and the Sun Server X4-2L was used for the database tier.

  • The 2-socket Sun Server X4-2 demonstrated 16% better performance when compared to the 2-socket IBM X3650 M4 server result of 9,696.43 SPECjEnterprise2010 EjOPS.

  • This result used Oracle WebLogic Server 12c, Java HotSpot(TM) 64-Bit Server 1.7.0_40, Oracle Database 12c, and Oracle Linux.

Performance Landscape

Complete benchmark results are at the SPEC website, SPECjEnterprise2010 Results. The table below shows the top single application server, two-chip x86 results.

SPECjEnterprise2010 Performance Chart
as of 9/22/2013
Submitter EjOPS* Application Server Database Server
Oracle 11,259.88 1x Sun Server X4-2
2x 2.7 GHz Intel Xeon E5-2697 v2
Oracle WebLogic 12c (12.1.2)
1x Sun Server X4-2L
2x 2.7 GHz Intel Xeon E5-2697 v2
Oracle Database 12c (12.1.0.1)
IBM 9,696.43 1x IBM X3650 M4
2x 2.9 GHz Intel Xeon E5-2690
WebSphere Application Server V8.5
1x IBM X3650 M4
2x 2.9 GHz Intel Xeon E5-2690
IBM DB2 10.1
Oracle 8,310.19 1x Sun Server X3-2
2x 2.9 GHz Intel Xeon E5-2690
Oracle WebLogic 11g (10.3.6)
1x Sun Server X3-2L
2x 2.9 GHz Intel Xeon E5-2690
Oracle Database 11g (11.2.0.3)

* SPECjEnterprise2010 EjOPS, bigger is better.

Configuration Summary

Application Server:

1 x Sun Server X4-2
2 x 2.7 GHz Intel Xeon processor E5-2697 v2
256 GB memory
4 x 10 GbE NIC
Oracle Linux 5 Update 9 (kernel-2.6.39-400.124.1.el5uek)
Oracle WebLogic Server 12c (12.1.2)
Java HotSpot(TM) 64-Bit Server VM on Linux, version 1.7.0_40 (Java SE 7 Update 40)

Database Server:

1 x Sun Server X4-2L
2 x 2.7 GHz Intel Xeon E5-2697 v2
256 GB memory
1 x 10 GbE NIC
2 x FC HBA
3 x Sun StorageTek 2540 M2
Oracle Linux 5 Update 9 (kernel-2.6.39-400.124.1.el5uek)
Oracle Database 12c Enterprise Edition Release 12.1.0.1

Benchmark Description

SPECjEnterprise2010 is the third generation of the SPEC organization's J2EE end-to-end industry standard benchmark application. The SPECjEnterprise2010 benchmark has been designed and developed to cover the Java EE 5 specification's significantly expanded and simplified programming model, highlighting the major features used by developers in the industry today. This provides a real world workload driving the Application Server's implementation of the Java EE specification to its maximum potential and allowing maximum stressing of the underlying hardware and software systems.

The workload consists of an end to end web based order processing domain, an RMI and Web Services driven manufacturing domain and a supply chain model utilizing document based Web Services. The application is a collection of Java classes, Java Servlets, Java Server Pages, Enterprise Java Beans, Java Persistence Entities (pojo's) and Message Driven Beans.

The SPECjEnterprise2010 benchmark heavily exercises all parts of the underlying infrastructure that make up the application environment, including hardware, JVM software, database software, JDBC drivers, and the system network.

The primary metric of the SPECjEnterprise2010 benchmark is jEnterprise Operations Per Second ("SPECjEnterprise2010 EjOPS"). This metric is calculated by adding the metrics of the Dealership Management Application in the Dealer Domain and the Manufacturing Application in the Manufacturing Domain. There is no price/performance metric in this benchmark.

Key Points and Best Practices

  • Four Oracle WebLogic server instances were started using numactl binding 2 instances per chip.
  • Two Oracle database listener processes were started and each was bound to a separate chip.
  • Additional tuning information is in the report at http://spec.org.

See Also

Disclosure Statement

SPEC and the benchmark name SPECjEnterprise are registered trademarks of the Standard Performance Evaluation Corporation. Sun Server X4-2, 11,259.88 SPECjEnterprise2010 EjOPS; Sun Server X3-2, 8,310.19 SPECjEnterprise2010 EjOPS; IBM System X3650 M4, 9,696.43 SPECjEnterprise2010 EjOPS. Results from www.spec.org 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
« September 2013 »
SunMonTueWedThuFriSat
1
2
3
4
5
6
7
8
11
12
13
14
15
16
17
19
20
21
22
24
27
28
29
30
     
       
Today