Wednesday Nov 30, 2011

SPARC T4-4 Beats 8-CPU IBM POWER7 on TPC-H @3000GB Benchmark

Oracle's SPARC T4-4 server delivered a world record TPC-H @3000GB benchmark result for systems with four processors. This result beats eight processor results from IBM (POWER7) and HP (x86). The SPARC T4-4 server also delivered better performance per core than these eight processor systems from IBM and HP. Comparisons below are based upon system to system comparisons, highlighting Oracle's complete software and hardware solution.

This database world record result used Oracle's Sun Storage 2540-M2 arrays (rotating disk) connected to a SPARC T4-4 server running Oracle Solaris 11 and Oracle Database 11g Release 2 demonstrating the power of Oracle's integrated hardware and software solution.

  • The SPARC T4-4 server based configuration achieved a TPC-H scale factor 3000 world record for four processor systems of 205,792 QphH@3000GB with price/performance of $4.10/QphH@3000GB.

  • The SPARC T4-4 server with four SPARC T4 processors (total of 32 cores) is 7% faster than the IBM Power 780 server with eight POWER7 processors (total of 32 cores) on the TPC-H @3000GB benchmark.

  • The SPARC T4-4 server is 36% better in price performance compared to the IBM Power 780 server on the TPC-H @3000GB Benchmark.

  • The SPARC T4-4 server is 29% faster than the IBM Power 780 for data loading.

  • The SPARC T4-4 server is up to 3.4 times faster than the IBM Power 780 server for the Refresh Function.

  • The SPARC T4-4 server with four SPARC T4 processors is 27% faster than the HP ProLiant DL980 G7 server with eight x86 processors on the TPC-H @3000GB benchmark.

  • The SPARC T4-4 server is 52% faster than the HP ProLiant DL980 G7 server for data loading.

  • The SPARC T4-4 server is up to 3.2 times faster than the HP ProLiant DL980 G7 for the Refresh Function.

  • The SPARC T4-4 server achieved a peak IO rate from the Oracle database of 17 GB/sec. This rate was independent of the storage used, as demonstrated by the TPC-H @3000TB benchmark which used twelve Sun Storage 2540-M2 arrays (rotating disk) and the TPC-H @1000TB benchmark which used four Sun Storage F5100 Flash Array devices (flash storage). [*]

  • The SPARC T4-4 server showed linear scaling from TPC-H @1000GB to TPC-H @3000GB. This demonstrates that the SPARC T4-4 server can handle the increasingly larger databases required of DSS systems. [*]

  • The SPARC T4-4 server benchmark results demonstrate a complete solution of building Decision Support Systems including data loading, business questions and refreshing data. Each phase usually has a time constraint and the SPARC T4-4 server shows superior performance during each phase.

[*] The TPC believes that comparisons of results published with different scale factors are misleading and discourages such comparisons.

Performance Landscape

The table lists the leading TPC-H @3000GB results for non-clustered systems.

TPC-H @3000GB, Non-Clustered Systems
System
Processor
P/C/T – Memory
Composite
(QphH)
$/perf
($/QphH)
Power
(QppH)
Throughput
(QthH)
Database Available
SPARC Enterprise M9000
3.0 GHz SPARC64 VII+
64/256/256 – 1024 GB
386,478.3 $18.19 316,835.8 471,428.6 Oracle 11g R2 09/22/11
SPARC T4-4
3.0 GHz SPARC T4
4/32/256 – 1024 GB
205,792.0 $4.10 190,325.1 222,515.9 Oracle 11g R2 05/31/12
SPARC Enterprise M9000
2.88 GHz SPARC64 VII
32/128/256 – 512 GB
198,907.5 $15.27 182,350.7 216,967.7 Oracle 11g R2 12/09/10
IBM Power 780
4.1 GHz POWER7
8/32/128 – 1024 GB
192,001.1 $6.37 210,368.4 175,237.4 Sybase 15.4 11/30/11
HP ProLiant DL980 G7
2.27 GHz Intel Xeon X7560
8/64/128 – 512 GB
162,601.7 $2.68 185,297.7 142,685.6 SQL Server 2008 10/13/10

P/C/T = Processors, Cores, Threads
QphH = the Composite Metric (bigger is better)
$/QphH = the Price/Performance metric in USD (smaller is better)
QppH = the Power Numerical Quantity
QthH = the Throughput Numerical Quantity

The following table lists data load times and refresh function times during the power run.

TPC-H @3000GB, Non-Clustered Systems
Database Load & Database Refresh
System
Processor
Data Loading
(h:m:s)
T4
Advan
RF1
(sec)
T4
Advan
RF2
(sec)
T4
Advan
SPARC T4-4
3.0 GHz SPARC T4
04:08:29 1.0x 67.1 1.0x 39.5 1.0x
IBM Power 780
4.1 GHz POWER7
05:51:50 1.5x 147.3 2.2x 133.2 3.4x
HP ProLiant DL980 G7
2.27 GHz Intel Xeon X7560
08:35:17 2.1x 173.0 2.6x 126.3 3.2x

Data Loading = database load time
RF1 = power test first refresh transaction
RF2 = power test second refresh transaction
T4 Advan = the ratio of time to T4 time

Complete benchmark results found at the TPC benchmark website http://www.tpc.org.

Configuration Summary and Results

Hardware Configuration:

SPARC T4-4 server
4 x SPARC T4 3.0 GHz processors (total of 32 cores, 128 threads)
1024 GB memory
8 x internal SAS (8 x 300 GB) disk drives

External Storage:

12 x Sun Storage 2540-M2 array storage, each with
12 x 15K RPM 300 GB drives, 2 controllers, 2 GB cache

Software Configuration:

Oracle Solaris 11 11/11
Oracle Database 11g Release 2 Enterprise Edition

Audited Results:

Database Size: 3000 GB (Scale Factor 3000)
TPC-H Composite: 205,792.0 QphH@3000GB
Price/performance: $4.10/QphH@3000GB
Available: 05/31/2012
Total 3 year Cost: $843,656
TPC-H Power: 190,325.1
TPC-H Throughput: 222,515.9
Database Load Time: 4:08:29

Benchmark Description

The TPC-H benchmark is a performance benchmark established by the Transaction Processing Council (TPC) to demonstrate Data Warehousing/Decision Support Systems (DSS). TPC-H measurements are produced for customers to evaluate the performance of various DSS systems. These queries and updates are executed against a standard database under controlled conditions. Performance projections and comparisons between different TPC-H Database sizes (100GB, 300GB, 1000GB, 3000GB, 10000GB, 30000GB and 100000GB) are not allowed by the TPC.

TPC-H is a data warehousing-oriented, non-industry-specific benchmark that consists of a large number of complex queries typical of decision support applications. It also includes some insert and delete activity that is intended to simulate loading and purging data from a warehouse. TPC-H measures the combined performance of a particular database manager on a specific computer system.

The main performance metric reported by TPC-H is called the TPC-H Composite Query-per-Hour Performance Metric (QphH@SF, where SF is the number of GB of raw data, referred to as the scale factor). QphH@SF is intended to summarize the ability of the system to process queries in both single and multiple user modes. The benchmark requires reporting of price/performance, which is the ratio of the total HW/SW cost plus 3 years maintenance to the QphH. A secondary metric is the storage efficiency, which is the ratio of total configured disk space in GB to the scale factor.

Key Points and Best Practices

  • Twelve Sun Storage 2540-M2 arrays were used for the benchmark. Each Sun Storage 2540-M2 array contains 12 15K RPM drives and is connected to a single dual port 8Gb FC HBA using 2 ports. Each Sun Storage 2540-M2 array showed 1.5 GB/sec for sequential read operations and showed linear scaling, achieving 18 GB/sec with twelve Sun Storage 2540-M2 arrays. These were stand alone IO tests.

  • The peak IO rate measured from the Oracle database was 17 GB/sec.

  • Oracle Solaris 11 11/11 required very little system tuning.

  • Some vendors try to make the point that storage ratios are of customer concern. However, storage ratio size has more to do with disk layout and the increasing capacities of disks – so this is not an important metric in which to compare systems.

  • The SPARC T4-4 server and Oracle Solaris efficiently managed the system load of over one thousand Oracle Database parallel processes.

  • Six Sun Storage 2540-M2 arrays were mirrored to another six Sun Storage 2540-M2 arrays on which all of the Oracle database files were placed. IO performance was high and balanced across all the arrays.

  • The TPC-H Refresh Function (RF) simulates periodical refresh portion of Data Warehouse by adding new sales and deleting old sales data. Parallel DML (parallel insert and delete in this case) and database log performance are a key for this function and the SPARC T4-4 server outperformed both the IBM POWER7 server and HP ProLiant DL980 G7 server. (See the RF columns above.)

See Also

Disclosure Statement

TPC-H, QphH, $/QphH are trademarks of Transaction Processing Performance Council (TPC). For more information, see www.tpc.org. SPARC T4-4 205,792.0 QphH@3000GB, $4.10/QphH@3000GB, available 5/31/12, 4 processors, 32 cores, 256 threads; IBM Power 780 QphH@3000GB, 192,001.1 QphH@3000GB, $6.37/QphH@3000GB, available 11/30/11, 8 processors, 32 cores, 128 threads; HP ProLiant DL980 G7 162,601.7 QphH@3000GB, $2.68/QphH@3000GB available 10/13/10, 8 processors, 64 cores, 128 threads.

Monday Oct 03, 2011

SPARC T4-4 Servers Set World Record on SPECjEnterprise2010, Beats IBM POWER7, Cisco x86

Oracle produced a world record SPECjEnterprise2010 benchmark result of 40,104.86 SPECjEnterprise2010 EjOPS using four of Oracle's SPARC T4-4 servers in the application tier and two more SPARC T4-4 servers for the database server.

  • The four SPARC T4-4 server configuration (sixteen SPARC T4 processors total, 3.0 GHz) demonstrated 2.4x better performance compared to the IBM Power 780 server (eight POWER7 processors, 3.86 THz) result of 16,646.34 SPECjEnterprise2010 EjOPS.

  • In the database tier, two SPARC T4-4 servers with a total of eight SPARC T4 processors at 3.0 GHz, processed 2.4x more transactions compared to the IBM result of 16,646.34 SPECjEnterprise2010 EjOPS which used four POWER7 processors at 3.55 GHz.

  • The four SPARC T4-4 server configuration demonstrated 1.5x better performance compared to the Cisco UCS B440 M2 Blade Server result of 26,118.67 SPECjEnterprise2010 EjOPS.

  • The four SPARC T4-4 server configuration demonstrated 2.3x better performance compared to the Cisco UCS B440 M1 Blade Server result of 17,301.86 SPECjEnterprise2010 EjOPS.

  • This result demonstrated less than 1 second average response times for all SPECjEnterprise2010 transactions and 90% of all transaction times took less than 1 second.

  • This result demonstrated a sustained Java EE 5 transaction load generated by approximately 320,000 users.

  • This result using 16 Oracle WebLogic 10.3.5 server instances demonstrated 4.8x better performance per application server instance when compared to the IBM result which used 32 WebSphere instances.

  • The SPARC T4-4 servers delivered a 6.7x price/performance advantage over the IBM Power 780 for the servers used in the application tier (see disclosure statement below for details). This price/performance advantage in the application tier was accomplished with a SPARC T4-4 server configuration with 2 TB of total memory compared to the IBM solution with 0.5 TB of memory.

  • The SPARC T4-4 servers had a 1.9x advantage over IBM in performance per space for the application tier (see disclosure statement below for details) even though the Oracle solution had four servers.

  • The four SPARC T4-4 servers used for the application tier used Oracle Solaris Containers to consolidate four Oracle WebLogic application server instances on each server to achieve this result.

  • The two SPARC T4-4 servers used for the database tier hosted Oracle Database 11g Release 2 and Oracle RAC cluster software using Oracle Automatic Storage Management (ASM).

  • Oracle Fusion Middleware provides a family of complete, integrated, hot pluggable and best-of-breed products known for enabling enterprise customers to create and run agile and intelligent business applications. Oracle WebLogic Server's on-going, record-setting Java application server performance demonstrates why so many customers rely on Oracle Fusion Middleware as their foundation for innovation.

Performance Landscape

Complete benchmark results are at the SPEC website, SPECjEnterprise2010 Results.

SPECjEnterprise2010 Performance Chart
as of 10/11/2011
Submitter EjOPS* Java EE Server DB Server
Oracle 40,104.86 4 x SPARC T4-4
4 chips, 32 cores, 3.0 GHz SPARC T4
Oracle WebLogic 11g (10.3.5)
2 x SPARC T4-4
4 chips, 32 cores, 3.0 GHz SPARC T4
Oracle 11g DB 11.2.0.2
Cisco 26,118.67 2 x Cisco UCS B440 M2
4 chips, 40 cores, 2.4 GHz Xeon E7-4870
Oracle WebLogic 11g (10.3.5)
1 x Cisco UCS C460 M2
4 chips, 40 cores, 2.4 GHz Xeon E7-4870
Oracle 11g DB 11.2.0.2
Cisco 17,301.86 2 x Cisco UCS B440 M1
4 chips, 32 cores, 2.26 GHz Xeon X7560
Oracle WebLogic 10.3.4
1 x Cisco UCS C460 M1
4 chips, 32 cores, 2.26 GHz Xeon X7560
Oracle 11g DB 11.2.0.2
IBM 16,646.34 1 x IBM Power 780
8 chips, 64 cores, 3.86 GHz POWER7
WebSphere Application Server V7.0
1 x IBM Power 750 Express
4 chips, 32 cores, 3.55 GHz POWER7
IBM DB2 Universal Database 9.7

* SPECjEnterprise2010 EjOPS (bigger is better)

Configuration Summary

Application Servers:

4 x SPARC T4-4 servers, each with
4 x 3.0 GHz SPARC T4 processors
512 GB memory
2 x 10GbE NIC
Oracle Solaris 10 8/11
Oracle WebLogic Server 11g Release 1 (10.3.5)
Java HotSpot(TM) 64-Bit Server VM on Solaris, version 1.6.0_26 (Java SE 6 Update 26)

Database Servers:

2 x SPARC T4-4 servers, each with
4 x 3.0 GHz SPARC T4 processors
1024 GB memory
2 x 10GbE NIC
4 x 8Gb FC HBA
Oracle Solaris 10 8/11
Oracle Database 11g Enterprise Edition Release 11.2.0.2
Oracle Real Application Clusters 11g Release 2

Storage Servers:

8 x Sun Fire X4270 M2 (12-Drive)
1 x 3.0 GHz Intel Xeon
8 GB memory
1 x 8Gb FC HBA
Oracle Solaris 11 Express 2010.11
8 x Sun Storage F5100 Flash Arrays

Switch Hardware:

2 x Sun Network 10GbE 72-port Top of Rack (ToR) Switch
1 x Brocade 5300 80-port Fiber Channel Switch

Benchmark Description

SPECjEnterprise2010 is the third generation of the SPEC organization's J2EE end-to-end industry standard benchmark application. The new SPECjEnterprise2010 benchmark has been re-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 web container, servlets, and web services
  • The EJB container
  • JPA 1.0 Persistence Model
  • JMS and Message Driven Beans
  • Transaction management
  • Database connectivity
Moreover, SPECjEnterprise2010 also 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). The primary metric for the SPECjEnterprise2010 benchmark 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 on each SPARC T4-4 server were hosted in 4 separate Oracle Solaris Containers to demonstrate consolidation of multiple application servers.
  • Each Oracle Solaris Container was bound to a separate processor set, each contained 7 cores (total 56 threads). This was done to improve performance by reducing memory access latency by using the physical memory closest to the processors. The default set was used for network and disk interrupt handling.
  • The Oracle WebLogic application servers were executed in the FX scheduling class to improve performance by reducing the frequency of context switches.
  • The Oracle database processes were run in 2 processor sets using psrset(1M) and executed in the FX scheduling class. This improved performance by reducing memory access latency and reducing context switches.
  • The Oracle log writer process was run in a separate processor set containing 2 threads and run in the RT scheduling class. This insured that the log writer had the most efficient use of CPU resources.

See Also

Disclosure Statement

SPEC and the benchmark name SPECjEnterprise are registered trademarks of the Standard Performance Evaluation Corporation. Results from www.spec.org as of 10/11/2011. SPARC T4-4, 40,104.86 SPECjEnterprise2010 EjOPS; Cisco UCS B440 M2, 26,118.67 SPECjEnterprise2010 EjOPS; Cisco UCS B440 M1, 17,301.86 SPECjEnterprise2010 EjOPS; IBM Power 780, 16,646.34 SPECjEnterprise2010 EjOPS.

SPECjEnterprise2010 models contemporary Java-based applications that run on large Java EE (Java Enterprise Edition) servers, backed by network infrastructure and database servers. Focusing on the critical Java EE server hardware & OS, the IBM result includes a Java EE server with a list price of $1.30 million. The Oracle Java EE servers have a list price of $0.47 million. The Java EE server price versus delivered EjOPS is $77.97/EjOPS for IBM versus $11.67/EjOPS for Oracle. Oracle's $/perf advantage is 6.7x better than IBM ($77.97/$11.67).

Pricing details for IBM, IBM p780 512GB based on public pricing at http://tpc.org/results/FDR/TPCH/TPC-H_1TB_IBM780_Sybase-FDR.pdf. Adjusted hardware costs to license all 64 cores. AIX pricing at: http://www-304.ibm.com/easyaccess3/fileserve?contentid=214347 and AIX Standard Edition V7.1 per processor (5765-G98-0017 64*2,600=$166,400). This gives application tier hardware & OS Price/perf: $77.97/EjOPS (1297956/16646.34)

Pricing details for Oracle, four SPARC T4-4 512 GB, HW acquisition price from Oracle's price list: $467,856 http://www.oracle.com. This gives application tier hardware & OS Price/perf: $11.67/EjOPS (467856/40104.86)

The Oracle application tier servers occupy 20U of space, 40,140.86/20=2005 EjOPS/U. The IBM application tier server occupies 16U of space, 16,646.34/16=1040 EjOPS/U. 2005/1040=1.9x

SPARC T4-4 Beats IBM POWER7 and HP Itanium on TPC-H @1000GB Benchmark

Oracle's SPARC T4-4 server configured with SPARC-T4 processors, Oracle's Sun Storage F5100 Flash Array storage, Oracle Solaris, and Oracle Database 11g Release 2 achieved a TPC-H benchmark performance result of 201,487 QphH@1000GB with price/performance of $4.60/QphH@1000GB.

  • The SPARC T4-4 server benchmark results demonstrate a complete solution of building Decision Support Systems including data loading, business questions and refreshing data. Each phase usually has a time constraint and the SPARC T4-4 server shows superior performance during each phase.

  • The SPARC T4-4 server is 22% faster than the 8-socket IBM POWER7 server with the same number of cores. The SPARC T4-4 server has over twice the performance per socket compared to the IBM POWER7 server.

  • The SPARC T4-4 server achieves 33% better price/performance than the IBM POWER7 server.

  • The SPARC T4-4 server is up to 4 times faster than the IBM POWER7 server for the Refresh Function.

  • The SPARC T4-4 server is 44% faster than the HP Superdome 2 server. The SPARC T4-4 server has 5.7x the performance per socket of the HP Superdome 2 server.

  • The SPARC T4-4 server is 62% better on price/performance than the HP Itanium server.

  • The SPARC T4-4 server is up to 3.7 times faster than the HP Itanium server for the Refresh Function.

  • The SPARC T4-4 server delivers nearly the same performance as Oracle's SPARC Enterprise M8000 server, but with 52% better price/performance on the TPC-H @1000GB benchmark.

  • Oracle used Storage Redundancy Level 3 as defined by the TPC-H 2.14.2 specification which is the strictest level.

  • This TPC-H result demonstrates that the SPARC T4-4 server can deliver the performance while running the increasingly larger databases required of DSS systems. The server measured more than 16 GB/sec of IO throughput through Oracle Database 11g Release 2 software while maintaining the high cpu load.

Performance Landscape

The table below lists published non-cluster results from comparable enterprise class systems from Oracle, IBM and HP. Each system was configured with 512 GB of memory.

TPC-H @1000GB

System
CPU type
Proc/Core/Thread
Composite
(QphH)
$/perf
($/QphH)
Power
(QppH)
Throughput
(QthH)
Database Available
SPARC Enterprise M8000
3 GHz SPARC64 VII+
16 / 64 / 128
209,533.6 $9.53 177,845.9 246,867.2 Oracle 11g 09/22/11
SPARC T4-4
3 GHz SPARC-T4
4 / 32 / 256
201,487.0 $4.60 181,760.6 223,354.2 Oracle 11g 10/30/11
IBM Power 780
4.14 GHz POWER7
8 / 32 / 128
164,747.2 $6.85 170,206.4 159,463.1 Sybase 03/31/11
HP Superdome 2
1.73 GHz Intel Itanium 9350
16 / 64 / 64
140,181.1 $12.15 139,181.0 141,188.3 Oracle 11g 10/20/10

QphH = the Composite Metric (bigger is better)
$/QphH = the Price/Performance metric (smaller is better)
QppH = the Power Numerical Quantity
QthH = the Throughput Numerical Quantity

Complete benchmark results found at the TPC benchmark website http://www.tpc.org.

Configuration Summary and Results

Hardware Configuration:

SPARC T4-4 server
4 x SPARC-T4 3.0 GHz processors (total of 32 cores, 128 threads)
512 GB memory
8 x internal SAS (8 x 300 GB) disk drives

External Storage:

4 x Sun Storage F5100 Flash Array storage, each with
80 x 24 GB Flash Modules

Software Configuration:

Oracle Solaris 10 8/11
Oracle Database 11g Release 2 Enterprise Edition

Audited Results:

Database Size: 1000 GB (Scale Factor 1000)
TPC-H Composite: 201,487 QphH@1000GB
Price/performance: $4.60/QphH@1000GB
Available: 10/30/2011
Total 3 Year Cost: $925,525
TPC-H Power: 181,760.6
TPC-H Throughput: 223,354.2
Database Load Time: 1:22:39

Benchmark Description

The TPC-H benchmark is a performance benchmark established by the Transaction Processing Council (TPC) to demonstrate Data Warehousing/Decision Support Systems (DSS). TPC-H measurements are produced for customers to evaluate the performance of various DSS systems. These queries and updates are executed against a standard database under controlled conditions. Performance projections and comparisons between different TPC-H Database sizes (100GB, 300GB, 1000GB, 3000GB and 10000GB) are not allowed by the TPC.

TPC-H is a data warehousing-oriented, non-industry-specific benchmark that consists of a large number of complex queries typical of decision support applications. It also includes some insert and delete activity that is intended to simulate loading and purging data from a warehouse. TPC-H measures the combined performance of a particular database manager on a specific computer system.

The main performance metric reported by TPC-H is called the TPC-H Composite Query-per-Hour Performance Metric (QphH@SF, where SF is the number of GB of raw data, referred to as the scale factor). QphH@SF is intended to summarize the ability of the system to process queries in both single and multi user modes. The benchmark requires reporting of price/performance, which is the ratio of QphH to total HW/SW cost plus 3 years maintenance.

Key Points and Best Practices

  • Four Sun Storage F5100 Flash Array devices were used for the benchmark. Each F5100 device contains 80 flash modules (FMODs). Twenty (20) FMODs from each F5100 device were connected to a single SAS 6 Gb HBA. A single F5100 device showed 4.16 GB/sec for sequential read and demonstrated linear scaling of 16.62 GB/sec with 4 x F5100 devices.

  • The IO rate from the Oracle database was over 16 GB/sec.

  • Oracle Solaris 10 8/11 required very little system tuning.

  • The SPARC T4-4 server and Oracle Solaris efficiently managed the system load of over one thousand Oracle parallel processes.

  • The Oracle database files for tables and indexes were managed by Oracle Automatic Storage Manager (ASM) with 4M stripe. Two F5100 devices were mirrored to another 2 F5100 devices under ASM. IO performance was high and balanced across all the FMODs.
  • The Oracle redo log files were mirrored across the F5100 devices using Oracle Solaris Volume Manager with 128K stripe.
  • Parallel degree on tables and indexes was set to 128. This setting worked the best for performance.
  • TPC-H Refresh Function simulates periodical Refresh portion of Data Warehouse by adding new sales and deleting old sales data. Parallel DML (parallel insert and delete in this case) and database log performance are a key for this function and the SPARC T4-4 server outperformed both HP Superdome 2 and IBM POWER7 servers.

See Also

Disclosure Statement

TPC-H, QphH, $/QphH are trademarks of Transaction Processing Performance Council (TPC). For more information, see www.tpc.org. SPARC T4-4 201,487 QphH@1000GB, $4.60/QphH@1000GB, avail 10/30/2011, 4 processors, 32 cores, 256 threads; SPARC Enterprise M8000 209,533.6 QphH@1000GB, $9.53/QphH@1000GB, avail 09/22/11, 16 processors, 64 cores, 128 threads; IBM Power 780 QphH@1000GB, 164,747.2 QphH@1000GB, $6.85/QphH@1000GB, avail 03/31/11, 8 processors, 32 cores, 128 threads; HP Integrity Superdome 2 140,181.1 QphH@1000GB, $12.15/QphH@1000GB avail 10/20/10, 16 processors, 64, cores, 64 threads.

Thursday Sep 29, 2011

SPARC T4 Processor Outperforms IBM POWER7 and Intel (Westmere AES-NI) on OpenSSL AES Encryption Test

Oracle's SPARC T4 processor is faster than the Intel Xeon X5690 (with AES-NI) and the IBM POWER7.

  • On single-thread OpenSSL encryption, the 2.85 GHz SPARC T4 processor is 4.3 times faster than the 3.5 GHz IBM POWER7 processor.

  • On single-thread OpenSSL encryption, the 2.85 GHz SPARC T4 processor is 17% faster than the 3.46 GHz Intel Xeon X5690 processor.

The SPARC T4 processor has Encryption Instruction Accelerators for encryption and decryption for AES and many other ciphers. The Intel Xeon X5690 processor has AES-NI instructions which accelerate only AES ciphers. The IBM POWER7 does not have cryptographic instructions, but cryptographic coprocessors are available.

Performance Landscape

The table below shows results when running the OpenSSL speed command with the AES-256-CBC cipher. The reported results are for a message size of 8192 bytes. Results are reported for a single thread and for running on all available hardware threads (no over subscribing).

OpenSSL Performance with
AES-256-CBC Encryption
Processor Performance (MB/sec)
1 Thread Maximum Throughput
(at number of threads)
SPARC T4, 2.85 GHz 769 11,967 (64)
Intel Xeon X5690, 3.46 GHz 660 7,362 (12)
IBM POWER7, 3.5 GHz 179 2,860 (est*)

(est*) The performance of the IBM POWER7 is estimated at 16 times the rate of the single thread performance. The estimate is considered an upper bound on expected performance for this processor.

Configuration Summary

SPARC Configuration:

SPARC T4-1 server
1 x SPARC T4 processors, 2.85 GHz
64 GB memory
Oracle Solaris 11

Intel Configuration:

Sun Fire X4270 M2 server
1 x Intel Xeon X5690 processors, 3.46 GHz
24 GB memory
Oracle Solaris 11

Software Configuration:

OpenSSL 1.0.0.d
gcc 3.4.3

Benchmark Description

The in-memory SSL performance was measured with the openssl command. openssl has an option for measuring the speed of various ciphers and message sizes. The actual command used to measure the speed of AES-256-CBC was:

openssl speed -multi {number of threads} -evp aes-256-cbc

openssl runs for several minutes and measures the speed, in units of MB/sec, of the specified cipher for messages of sizes 16 bytes to 8192 bytes.

Key Points and Best Practices

  • The Encryption Instruction Accelerators are accessed through a platform independent API for cryptographic engines.
  • The OpenSSL libraries use the API. The default is to not use the Encryption Instruction Accelerators.
  • Cryptography is compute intensive. Using all available threads streams, both the SPARC T4 processor and the Intel Xeon processor were able to saturate the memory bandwidth of the respective systems.

See Also

Disclosure Statement

Copyright 2011, 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 9/26/2011.

Wednesday Sep 28, 2011

SPARC T4 Servers Set World Record on PeopleSoft HRMS 9.1

Oracle's SPARC T4-4 servers running Oracle's PeopleSoft HRMS Self-Service 9.1 benchmark and Oracle Database 11g Release 2 achieved World Record performance on Oracle Solaris 10.

  • Using two SPARC T4-4 servers to run the application and database tiers and one SPARC T4-2 server to run the webserver tier, Oracle demonstrated world record performance of 15,000 concurrent users running the PeopleSoft HRMS Self-Service 9.1 benchmark.

  • The combination of the SPARC T4 servers running the PeopleSoft HRMS 9.1 benchmark supports 3.8x more online users with faster response time compared to the best published result from IBM on the previous PeopleSoft HRMS 8.9 benchmark.

  • The average CPU utilization on the SPARC T4-4 server in the application tier handling 15,000 users was less than 50%, leaving significant room for application growth.

  • The SPARC T4-4 server on the application tier used Oracle Solaris Containers which provide a flexible, scalable and manageable virtualization environment.

Performance Landscape

PeopleSoft HRMS Self-Service 9.1 Benchmark
Systems Processors Users Ave Response -
Search (sec)
Ave Response -
Save (sec)
SPARC T4-2 (web)
SPARC T4-4 (app)
SPARC T4-4 (db)
2 x SPARC T4, 2.85 GHz
4 x SPARC T4, 3.0 GHz
4 x SPARC T4, 3.0 GHz
15,000 1.01 0.63
PeopleSoft HRMS Self-Service 8.9 Benchmark
IBM Power 570 (web/app)
IBM Power 570 (db)
12 x POWER5, 1.9 GHz
4 x POWER5, 1.9 GHz
4,000 1.74 1.25
IBM p690 (web)
IBM p690 (app)
IBM p690 (db)
4 x POWER4, 1.9 GHz
12 x POWER4, 1.9 GHz
6 x 4392 MPIS/Gen1
4,000 1.35 1.01

The main differences between version 9.1 and version 8.9 of the benchmark are:

  • the database expanded from 100K employees and 20K managers to 500K employees and 100K managers,
  • the manager data was expanded,
  • a new transaction, "Employee Add Profile," was added, the percent of users executing it is less then 2%, and the transaction has a heavier footprint,
  • version 9.1 has a different benchmark metric (Average Response search/save time for x number of users) versus single user search/save time,
  • newer versions of the PeopleSoft application and PeopleTools software are used.

Configuration Summary

Application Server:

1 x SPARC T4-4 server
4 x SPARC T4 processors 3.0 GHz
512 GB main memory
5 x 300 GB SAS internal disks,
2 x 100 GB internal SSDs
1 x 300 GB internal SSD
Oracle Solaris 10 8/11
PeopleSoft PeopleTools 8.51.02
PeopleSoft HCM 9.1
Oracle Tuxedo, Version 10.3.0.0, 64-bit, Patch Level 031
Java HotSpot(TM) 64-Bit Server VM on Solaris, version 1.6.0_20

Web Server:

1 x SPARC T4-2 server
2 x SPARC T4 processors 2.85 GHz
256 GB main memory
1 x 300 GB SAS internal disks
1 x 300 GB internal SSD
Oracle Solaris 10 8/11
PeopleSoft PeopleTools 8.51.02
Oracle WebLogic Server 11g (10.3.3)
Java HotSpot(TM) 64-Bit Server VM on Solaris, version 1.6.0_20

Database Server:

1 x SPARC T4-4 server
4 x SPARC T4 processors 3.0 GHz
256 GB main memory
3 x 300 GB SAS internal disks
1 x Sun Storage F5100 Flash Array (80 flash modules)
Oracle Solaris 10 8/11
Oracle Database 11g Release 2

Benchmark Description

The purpose of the PeopleSoft HRMS Self-Service 9.1 benchmark is to measure comparative online performance of the selected processes in PeopleSoft Enterprise HCM 9.1 with Oracle Database 11g. The benchmark kit is an Oracle standard benchmark kit run by all platform vendors to measure the performance. It's an OLTP benchmark with no dependency on remote COBOL calls, there is no batch workload, and DB SQLs are moderately complex. The results are certified by Oracle and a white paper is published.

PeopleSoft defines a business transaction as a series of HTML pages that guide a user through a particular scenario. Users are defined as corporate Employees, Managers and HR administrators. The benchmark consists of 14 scenarios which emulate users performing typical HCM transactions such as viewing paychecks, promoting and hiring employees, updating employee profiles and other typical HCM application transactions.

All these transactions are well-defined in the PeopleSoft HR Self-Service 9.1 benchmark kit. The benchmark metric is the Average Response Time for search and save for 15,000 users..

Key Points and Best Practices

  • The application tier was configured with two PeopleSoft application server instances on the SPARC T4-4 server hosted in two separate Oracle Solaris Containers to demonstrate consolidation of multiple application, ease of administration, and load balancing.

  • Each PeopleSoft Application Server instance running in an Oracle Solaris Container was configured to run 5 application server Domains with 30 application server instances to be able to effectively handle the 15,000 users workload with zero application server queuing and minimal use of resources.

  • The web tier was configured with 20 WebLogic instances and with 4 GB JVM heap size to load balance transactions across 10 PeopleSoft Domains. That enables equitable distribution of transactions and scaling to high number of users.

  • Internal SSDs were configured in the application tier to host PeopleSoft Application Servers object CACHE file systems and in the web tier for WebLogic servers' logging providing near zero millisecond service time and faster server response time.

See Also

Disclosure Statement

Oracle's PeopleSoft HRMS 9.1 benchmark, www.oracle.com/us/solutions/benchmark/apps-benchmark/peoplesoft-167486.html, results 9/26/2011.

Tuesday Sep 27, 2011

SPARC T4-4 Server Sets World Record on PeopleSoft Payroll (N.A.) 9.1, Outperforms IBM Mainframe, HP Itanium

Oracle's SPARC T4-4 server achieved world record performance on the Unicode version of Oracle's PeopleSoft Enterprise Payroll (N.A) 9.1 extra-large volume model benchmark using Oracle Database 11g Release 2 running on Oracle Solaris 10.

  • The SPARC T4-4 server was able to process 1,460,544 payments/hour using PeopleSoft Payroll N.A 9.1.

  • The SPARC T4-4 server UNICODE result of 30.84 minutes on Payroll 9.1 is 2.8x faster than IBM z10 EC 2097 Payroll 9.0 (UNICODE version) result of 87.4 minutes. The IBM mainframe is rated at 6,512 MIPS.

  • The SPARC T4-4 server UNICODE result of 30.84 minutes on Payroll 9.1 is 3.1x faster than HP rx7640 Itanium2 non-UNICODE result of 96.17 minutes, on Payroll 9.0.

  • The average CPU utilization on the SPARC T4-4 server was only 30%, leaving significant room for business growth.

  • The SPARC T4-4 server processed payroll for 500,000 employees, 750,000 payments, in 30.84 minutes compared to the earlier world record result of 46.76 minutes on Oracle's SPARC Enterprise M5000 server.

  • The SPARC Enterprise M5000 server configured with eight 2.66 GHz SPARC64 VII processors has a result of 46.76 minutes on Payroll 9.1. That is 7% better than the result of 50.11 minutes on the SPARC Enterprise M5000 server configured with eight 2.53 GHz SPARC64 VII processors on Payroll 9.0. The difference in clock speed between the two processors is ~5%. That is close to the difference in the two results, thereby showing that the impact of the Payroll 9.1 benchmark on the overall result is about the same as that of Payroll 9.0.

Performance Landscape

PeopleSoft Payroll (N.A.) 9.1 – 500K Employees (7 Million SQL PayCalc, Unicode)

System OS/Database Payroll Processing
Result (minutes)
Run 1
(minutes)
Num of
Streams
SPARC T4-4, 4 x 3.0 GHz SPARC T4 Solaris/Oracle 11g 30.84 43.76 96
SPARC M5000, 8 x 2.66 GHz SPARC64 VII+ Solaris/Oracle 11g 46.76 66.28 32

PeopleSoft Payroll (N.A.) 9.0 – 500K Employees (3 Million SQL PayCalc, Non-Unicode)

System OS/Database Time in Minutes Num of
Streams
Payroll
Processing
Result
Run 1 Run 2 Run 3
Sun M5000, 8 x 2.53 GHz SPARC64 VII Solaris/Oracle 11g 50.11 73.88 534.20 1267.06 32
IBM z10 EC 2097, 9 x 4.4 GHz Gen1 Z/OS /DB2 58.96 80.5 250.68 462.6 8
IBM z10 EC 2097, 9 x 4.4 GHz Gen1 Z/OS /DB2 87.4 ** 107.6 - - 8
HP rx7640, 8 x 1.6 GHz Itanium2 HP-UX/Oracle 11g 96.17 133.63 712.72 1665.01 32

** This result was run with Unicode. The IBM z10 EC 2097 UNICODE result of 87.4 minutes is 48% slower than IBM z10 EC 2097 non-UNICODE result of 58.96 minutes, both on Payroll 9.0, each configured with nine 4.4GHz Gen1 processors.

Payroll 9.1 Compared to Payroll 9.0

Please note that Payroll 9.1 is Unicode based and Payroll 9.0 had non-Unicode and Unicode versions of the workload. There are 7 million executions of an SQL statement for the PayCalc batch process in Payroll 9.1 and 3 million executions of the same SQL statement for the PayCalc batch process in Payroll 9.0. This gets reflected in the elapsed time (27.33 min for 9.1 and 23.78 min for 9.0). The elapsed times of all other batch processes is lower (better) on 9.1.

Configuration Summary

Hardware Configuration:

SPARC T4-4 server
4 x 3.0 GHz SPARC T4 processors
256 GB memory
Sun Storage F5100 Flash Array
80 x 24 GB FMODs

Software Configuration:

Oracle Solaris 10 8/11
PeopleSoft HRMS and Campus Solutions 9.10.303
PeopleSoft Enterprise (PeopleTools) 8.51.035
Oracle Database 11g Release 2 11.2.0.1 (64-bit)
Micro Focus COBOLServer Express 5.1 (64-bit)

Benchmark Description

The PeopleSoft 9.1 Payroll (North America) benchmark is a performance benchmark established by PeopleSoft to demonstrate system performance for a range of processing volumes in a specific configuration. This information may be used to determine the software, hardware, and network configurations necessary to support processing volumes. This workload represents large batch runs typical of OLTP workloads during a mass update.

To measure five application business process run times for a database representing a large organization. The five processes are:

  • Paysheet Creation: Generates payroll data worksheets consisting of standard payroll information for each employee for a given pay cycle.

  • Payroll Calculation: Looks at paysheets and calculates checks for those employees.

  • Payroll Confirmation: Takes information generated by Payroll Calculation and updates the employees' balances with the calculated amounts.

  • Print Advice forms: The process takes the information generated by Payroll Calculations and Confirmation and produces an Advice for each employee to report Earnings, Taxes, Deduction, etc.

  • Create Direct Deposit File: The process takes information generated by the above processes and produces an electronic transmittal file that is used to transfer payroll funds directly into an employee's bank account.

Key Points and Best Practices

  • The SPARC T4-4 server with the Sun Storage F5100 Flash Array device had an average read throughput of up to 103 MB/sec and an average write throughput of up to 124 MB/sec while consuming 30% CPU on average.

  • The Sun Storage F5100 Flash Array device is a solid-state device that provides a read latency of only 0.5 msec. That is about 10 times faster than the normal disk latencies of 5 msec measured on this benchmark.

See Also

  • Oracle PeopleSoft Benchmark White Papers
    oracle.com
  • PeopleSoft Enterprise Human Capital Management (Payroll)
    oracle.com

  • PeopleSoft Enterprise Payroll 9.1 Using Oracle for Solaris (Unicode) on an Oracle's SPARC T4-4 – White Paper
    oracle.com

  • SPARC T4-4 Server
    oracle.com
  • Oracle Solaris
    oracle.com
  • Oracle Database 11g Release 2 Enterprise Edition
    oracle.com
  • Sun Storage F5100 Flash Array
    oracle.com

Disclosure Statement

Oracle's PeopleSoft Payroll 9.1 benchmark, SPARC T4-4 30.84 min,
http://www.oracle.com/us/solutions/benchmark/apps-benchmark/peoplesoft-167486.html, results 9/26/2011.

Friday Aug 12, 2011

Sun Blade X6270 M2 with Oracle WebLogic World Record 2 Processor SPECjEnterprise 2010 Benchmark

Oracle produced a World Record single application server using 2 chips result for the SPECjEnterprise2010 benchmark of 5,427.42 SPECjEnterprise2010 EjOPS using one of Oracle's Sun Blade X6270 M2 server module for the application tier and one Sun Blade X6270 M2 server module for the database.

  • The Sun Blade X6270 M2 server module equipped with two Intel Xeon X5690 processors running at 3.46 GHz, demonstrated 47% better performance compared to the 2-chip IBM System HS22 server result of 3,694.35 SPECjEnterprise2010 EjOPS using the same model of Intel Xeon X5690 processor.

  • The Sun Blade X6270 M2 server module running the application tier demonstrated 33% better performance compared to the 2-chip IBM Power 730 Express server result of 4,062.38 SPECjEnterprise2010 EjOPS.

  • The Sun Blade X6270 M2 server modules used Oracle WebLogic Server 11g Release 1 (10.3.5) application, Java SE 6 Update 26, and Oracle Database 11g Release 2 to produce this result.

Performance Landscape

Complete benchmark results are at the SPEC website, SPECjEnterprise2010 Results.

SPECjEnterprise2010 Performance Chart
as of 8/11/2011
Submitter EjOPS* Application Server Database Server
Oracle 5,427.42 1x Sun Blade X6270 M2
2x 3.46 GHz Intel Xeon X5690
Oracle WebLogic 11g (10.3.5)
1x Sun Blade X6270 M2
2x 3.46 GHz Intel Xeon X5690
Oracle 11g DB 11.2.0.2
IBM 4,062.38 1x IBM Power 730 Express
2x 3.5 GHz POWER 7
WebSphere Application Server V7
1x IBM BladeCenter PS701
1x 3.0 GHz POWER 7
IBM DB2 9.7 Workgroup Server Edition FP3a
IBM 3,694.35 1x IBM HS22
2x 3.46 GHz Intel Xeon X5690
WebSphere Application Server V8
1x IBM x3850 X5
2x 2.4 GHz Intel Xeon E7-4870
IBM DB2 9.7 FP3a

* SPECjEnterprise2010 EjOPS, bigger is better.

Configuration Summary

Application Server:
    1 x Sun Blade X6270 M2
      2 x 3.46 GHz Intel Xeon X5690
      48 GB memory
      4 x 10 GbE NIC
      Oracle Linux 5 Update 6
      Oracle WebLogic Server 11g Release 1 (10.3.5)
      Java HotSpot(TM) 64-Bit Server VM on Linux, version 1.6.0_26 (Java SE 6 Update 26)

Database Server:

    1 x Sun Blade X6270 M2
      2 x 3.46 GHz Intel Xeon X5690
      144 GB memory
      2 x 10 GbE NIC
      2 x Sun Storage 6180
      Oracle Linux 5 Update 6
      Oracle Database 11g Enterprise Edition Release 11.2.0.2

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 5.0 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"). The primary metric for the SPECjEnterprise2010 benchmark 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

  • Two Oracle WebLogic server instances were started using numactl binding 1 instance 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 Blade X6270 M2, 5,427.42 SPECjEnterprise2010 EjOPS; IBM Power 730 Express, 4,062.38 SPECjEnterprise2010 EjOPS; IBM System HS22, 3,694.35 SPECjEnterprise2010 EjOPS. Results from www.spec.org as of 8/11/2011.

Friday Jun 03, 2011

SPARC Enterprise M8000 with Oracle 11g Beats IBM POWER7 on TPC-H @1000GB Benchmark

Oracle's SPARC Enterprise M8000 server configured with SPARC64 VII+ processors, Oracle's Sun Storage F5100 Flash Array storage, Oracle Solaris, and Oracle Database 11g Release 2 achieved a TPC-H performance result of 209,533 QphH@1000GB with price/performance of $9.53/QphH@1000GB.

Oracle's SPARC server surpasses the performance of the IBM POWER7 server on the 1 TB TPC-H decision support benchmark.

Oracle focuses on the performance of the complete hardware and software stack. Implementation details such as the number of cores or the number of threads obscures the important metric of delivered system performance. The SPARC Enterprise M8000 server delivers higher performance than the IBM Power 780 even though the SPARC VII+ processor-core is 1.6x slower than the POWER7 processor-core.

  • The SPARC Enterprise M8000 server is 27% faster than the IBM Power 780. IBM's reputed single-thread performance leadership does not provide benefit for throughput.

  • Oracle beats IBM Power with better performance. This shows that Oracle's focus on integrated system design provides more customer value than IBM's focus on per core performance.

  • The SPARC Enterprise M8000 server is up to 3.8 times faster than the IBM Power 780 for Refresh Function. Again, IBM's reputed single-thread performance leadership does not provide benefit for this important function.

  • The SPARC Enterprise M8000 server is 49% faster than the HP Superdome 2 (1.73 GHz Itanium 9350).

  • The SPARC Enterprise M8000 server is 22% better price performance than the HP Superdome 2 (1.73 GHz Itanium 9350).

  • The SPARC Enterprise M8000 server is 2 times faster than the HP Superdome 2 (1.73 GHz Itanium 9350) for Refresh Function.

  • Oracle used Storage Redundancy Level 3 as defined by the TPC-H 2.14.0 specification which is the highest level.

  • One should focus on the performance of the complete hardware and software stack since server implementation details such as the number of cores or the number of threads obscures the important metric of delivered system performance.

  • This TPC-H result demonstrates that the SPARC Enterprise M8000 server can handle the increasingly large databases required of DSS systems. The server delivered more than 16 GB/sec of IO throughput through Oracle Database 11g Release 2 software maintaining high cpu load.

Performance Landscape

The table below lists published results from comparable enterprise class systems from Oracle, HP and IBM. Each system was configured with 512 GB of memory.

TPC-H @1000GB

System
CPU type
Proc/Core/Thread
Composite
(QphH)
$/perf
($/QphH)
Power
(QppH)
Throughput
(QthH)
Database Available
SPARC Enterprise M8000
3 GHz SPARC64 VII+
16 / 64 / 128
209,533.6 $9.53 177,845.9 246,867.2 Oracle 11g 09/22/11
IBM Power 780
4.14 GHz POWER7
8 / 32 / 128
164,747.2 $6.85 170,206.4 159,463.1 Sybase 03/31/11
HP SuperDome 2
1.73 GHz Intel Itanium 9350
16 / 64 / 64
140,181.1 $12.15 139,181.0 141,188.3 Oracle 11g 10/20/10

QphH = the Composite Metric (bigger is better)
$/QphH = the Price/Performance metric (smaller is better)
QppH = the Power Numerical Quantity
QthH = the Throughput Numerical Quantity

Complete benchmark results found at the TPC benchmark website http://www.tpc.org.

Configuration Summary and Results

Server:

SPARC Enterprise M8000 server
16 x SPARC64 VII+ 3.0 GHz processors (total of 64 cores, 128 threads)
512 GB memory
12 x internal SAS (12 x 300 GB) disk drives

External Storage:

4 x Sun Storage F5100 Flash Array device, each with
80 x 24 GB Flash Modules

Software:

Oracle Solaris 10 8/11
Oracle Database 11g Release 2 Enterprise Edition

Audited Results:

Database Size: 1000 GB (Scale Factor 3000)
TPC-H Composite: 209,533.6 QphH@1000GB
Price/performance: $9.53/QphH@1000GB
Available: 09/22/2011
Total 3 year Cost: $1,995,715
TPC-H Power: 177,845.9
TPC-H Throughput: 246,867.2
Database Load Time: 1:27:12

Benchmark Description

The TPC-H benchmark is a performance benchmark established by the Transaction Processing Council (TPC) to demonstrate Data Warehousing/Decision Support Systems (DSS). TPC-H measurements are produced for customers to evaluate the performance of various DSS systems. These queries and updates are executed against a standard database under controlled conditions. Performance projections and comparisons between different TPC-H Database sizes (100GB, 300GB, 1000GB, 3000GB and 10000GB) are not allowed by the TPC.

TPC-H is a data warehousing-oriented, non-industry-specific benchmark that consists of a large number of complex queries typical of decision support applications. It also includes some insert and delete activity that is intended to simulate loading and purging data from a warehouse. TPC-H measures the combined performance of a particular database manager on a specific computer system.

The main performance metric reported by TPC-H is called the TPC-H Composite Query-per-Hour Performance Metric (QphH@SF, where SF is the number of GB of raw data, referred to as the scale factor). QphH@SF is intended to summarize the ability of the system to process queries in both single and multi user modes. The benchmark requires reporting of price/performance, which is the ratio of QphH to total HW/SW cost plus 3 years maintenance.

Key Points and Best Practices

  • Four Sun Storage F5100 Flash Array devices were used for the benchmark. Each F5100 device contains 80 Flash Modules (FMODs). Twenty (20) FMODs from each F5100 device were connected to a single SAS 6 Gb HBA. A single F5100 device showed 4.16 GB/sec for sequential read and demonstrated linear scaling of 16.62 GB/sec with 4 x F5100 devices.
  • The IO rate from the Oracle database was over 16 GB/sec.
  • Oracle Solaris 10 8/11 required very little system tuning.
  • The SPARC Enterprise M8000 server and Oracle Solaris efficiently managed the system load of over one thousand Oracle parallel processes.
  • The Oracle database files were mirrored under Solaris Volume Manager (SVM). Two F5100 arrays were mirrored to another 2 F5100 arrays. IO performance was good and balanced across all the FMODs. Because of the SVM mirror one of the durability tests, the disk/controller failure test, was transparent to the Oracle database.

See Also

Disclosure Statement

SPARC Enterprise M8000 209,533.6 QphH@1000GB, $9.53/QphH@1000GB, avail 09/22/11, IBM Power 780 QphH@1000GB, 164,747.2 QphH@1000GB, $6.85/QphH@1000GB, avail 03/31/11, HP Integrity Superdome 2 140,181.1 QphH@1000GB, $12.15/QphH@1000GB avail 10/20/10, TPC-H, QphH, $/QphH tm of Transaction Processing Performance Council (TPC). More info www.tpc.org.

Friday Mar 25, 2011

SPARC Enterprise M9000 with Oracle Database 11g Delivers World Record Single Server TPC-H @3000GB Result

Oracle's SPARC Enterprise M9000 server delivers single-system TPC-H @3000GB world record performance. The SPARC Enterprise M9000 server along with Oracle's Sun Storage 6180 arrays and running Oracle Database 11g Release 2 on the Oracle Solaris operating system proves the power of Oracle's integrated solution.

  • The SPARC Enterprise M9000 server configured with SPARC64 VII+ processors, Sun Storage 6180 arrays and running Oracle Solaris 10 combined with Oracle Database 11g Release 2 achieved World Record TPC-H performance of 386,478.3 QphH@3000GB for non-clustered systems.

  • The SPARC Enterprise M9000 server running the Oracle Database 11g Release 2 software is 2.5 times faster than the IBM p595 (POWER6) server which ran with Sybase IQ v.15.1 database software.

  • The SPARC Enterprise M9000 server is 3.4 times faster than the IBM p595 server for data loading.

  • The SPARC Enterprise M9000 server is 3.5 times faster than the IBM p595 server for Refresh Function.

  • The SPARC Enterprise M9000 server configured with Sun Storage 6180 arrays shows linear scaling up to the maximum delivered IO performance of 48.3 GB/sec as measured by vdbench.

  • The SPARC Enterprise M9000 server running the Oracle Database 11g Release 2 software is 2.4 times faster than the HP ProLiant DL980 server which used Microsoft SQL Server 2008 R2 Enterprise Edition software.

  • The SPARC Enterprise M9000 server is 2.9 times faster than the HP ProLiant DL980 server for data loading.

  • The SPARC Enterprise M9000 server is 4 times faster than the HP ProLiant DL980 server for Refresh Function.

  • A 1.94x improvement was delivered by the SPARC Enterprise M9000 server result using 64 SPARC64 VII+ processors compared to the previous Sun SPARC Enterprise M9000 server result which used 32 SPARC64 VII processes.

  • Oracle's TPC-H result shows that the SPARC Enterprise M9000 server can handle the increasingly large databases required of DSS systems. The IO rate as measured by the Oracle database is over 40 GB/sec.

  • Oracle used Storage Redundancy Level 3 as defined by the TPC-H 2.14.0 specification which is the highest level.

Performance Landscape

TPC-H @3000GB, Non-Clustered Systems

System
CPU type
Memory
Composite
(QphH)
$/perf
($/QphH)
Power
(QppH)
Throughput
(QthH)
Database Available
SPARC Enterprise M9000
3 GHz SPARC64 VII+
1024 GB
386,478.3 $18.19 316,835.8 471,428.6 Oracle 11g 09/22/11
Sun SPARC Enterprise M9000
2.88 GHz SPARC64 VII
512 GB
198,907.5 $15.27 182,350.7 216,967.7 Oracle 11g 12/09/10
HP ProLiant DL980 G7
2.27 GHz Intel Xeon X7560
512 GB
162,601.7 $2.68 185,297.7 142,601.7 SQL Server 10/13/10
IBM Power 595
5.0 GHz POWER6
512 GB
156,537.3 $20.60 142,790.7 171,607.4 Sybase 11/24/09

QphH = the Composite Metric (bigger is better)
$/QphH = the Price/Performance metric (smaller is better)
QppH = the Power Numerical Quantity
QthH = the Throughput Numerical Quantity

Complete benchmark results found at the TPC benchmark website http://www.tpc.org.

Configuration Summary and Results

Server:

SPARC Enterprise M9000
64 x SPARC VII+ 3.0 GHz processors
1024 GB memory
4 x internal SAS (4 x 146 GB)

External Storage:

32 x Sun Storage 6180 arrays (each with 16 x 600 GB)

Software:

Oracle Solaris 10 9/10
Oracle Database 11g Release 2 Enterprise Edition

Audited Results:

Database Size: 3000 GB (Scale Factor 3000)
TPC-H Composite: 386,478.3 QphH@3000GB
Price/performance: $18.19/QphH@3000GB
Available: 09/22/2011
Total 3 year Cost: $7,030,009
TPC-H Power: 316,835.8
TPC-H Throughput: 471,428.6
Database Load Time: 2:59:01

Benchmark Description

The TPC-H benchmark is a performance benchmark established by the Transaction Processing Council (TPC) to demonstrate Data Warehousing/Decision Support Systems (DSS). TPC-H measurements are produced for customers to evaluate the performance of various DSS systems. These queries and updates are executed against a standard database under controlled conditions. Performance projections and comparisons between different TPC-H Database sizes (100GB, 300GB, 1000GB, 3000GB and 10000GB) are not allowed by the TPC.

TPC-H is a data warehousing-oriented, non-industry-specific benchmark that consists of a large number of complex queries typical of decision support applications. It also includes some insert and delete activity that is intended to simulate loading and purging data from a warehouse. TPC-H measures the combined performance of a particular database manager on a specific computer system.

The main performance metric reported by TPC-H is called the TPC-H Composite Query-per-Hour Performance Metric (QphH@SF, where SF is the number of GB of raw data, referred to as the scale factor). QphH@SF is intended to summarize the ability of the system to process queries in both single and multi user modes. The benchmark requires reporting of price/performance, which is the ratio of QphH to total HW/SW cost plus 3 years maintenance.

Key Points and Best Practices

  • The Sun Storage 6180 array showed linear scalability of 48.3 GB/sec Sequential Read with thirty-two Sun Storage 6180 arrays. Scaling could continue if there are more arrays available.
  • Oracle Solaris 10 9/10 required very little system tuning.
  • The optimal Sun Storage 6180 arrays configuration for the benchmark was to set up 1 disk per volume instead of multiple disks per volume and let Oracle Oracle Automatic Storage Management (ASM) mirror. Presenting as many volumes as possible to Oracle database gave the highest scan rate.

  • The storage was managed by ASM with 4 MB stripe size. 1 MB is the default stripe size but 4 MB works better for large databases.

  • All the Oracle database files, except TEMP tablespace, were mirrored under ASM. 16 x Sun Storage 6180 arrays (256 disks) were mirrored to another 16 x Sun Storage 6180 arrays using ASM. IO performance was good and balanced across all the disks. With the ASM mirror the benchmark passed the ACID (Atomicity, Consistency, Isolation and Durablity) test.

  • Oracle database tables were 256-way partitioned. The parallel degree for each table was set to 256 to match the number of available cores. This setting worked the best for performance.

  • Oracle Database 11g Release 2 feature Automatic Parallel Degree Policy was set to AUTO for the benchmark. This enabled automatic degree of parallelism, statement queuing and in-memory parallel execution.

See Also

Disclosure Statement

SPARC Enterprise M9000 386,478.3 QphH@3000GB, $18.19/QphH@3000GB, avail 09/22/11, IBM Power 595 QphH@3000GB, 156,537.3 QphH@3000GB, $20.60/QphH@3000GB, avail 11/24/09, HP ProLiant DL980 G7 162,601.7 QphH@3000GB, $2.68/QphH@3000GB avail 10/13/10, TPC-H, QphH, $/QphH tm of Transaction Processing Performance Council (TPC). More info www.tpc.org.

Tuesday Mar 22, 2011

Netra SPARC T3-1 22% Faster Than IBM Running Oracle Communications ASAP

Oracle's Netra SPARC T3-1 server delivered better performance than the IBM Power 570 server running the Oracle Communications ASAP application. Oracle Communications ASAP is used by the world's leading communication providers to enable voice, data, video and content services across wireless, wireline and satellite networks.

  • A Netra SPARC T3-1 server is 22% faster than the IBM Power 570 server delivering higher order volume throughput. This was achieved by consolidating Oracle Database 11g Release 2 and Oracle Communications ASAP 7.0.2 software onto a single Netra SPARC T3-1 server.

  • Oracle's Netra servers are NEBS level 3 certified, unlike the competition. NEBS is a set of safety, physical, and environmental design guidelines for telecommunications equipment in the United States.

  • A single Netra SPARC T3-1 server takes one-eighth the rack space of an IBM Power 570 system.

  • The single processor Netra SPARC T3-1 server beat an eight processor IBM Power 570 server.

  • The ASAP result which was run on the Netra SPARC T3-1 server is the highest single-system throughput ever measured for this benchmark.

Performance Landscape

Results of Oracle Communications ASAP run with Oracle Database 11g.

System Processor Memory OS Orders/hour Version
Netra SPARC T3-1 1 x 1.65 GHz SPARC T3 128 GB Solaris 10 570,000 7.0.2
IBM Power 570 8 x 5 GHz POWER6 128 GB AIX 6.1.2 463,500 7.0

In both cases, server utilization ranged between 60 and 75%.

Configuration Summary

Hardware Configuration:

Netra SPARC T3-1
1 x 1.65 GHz T3 processor
128 GB memory
Sun Storage 7410 Unified Storage System with one Sun Storage J4400 array

Software Configuration:

Oracle Solaris 10 9/10
Oracle Database 11g Release 2 (11.2.0.1.0)
Java Platform, Standard Edition 6 Update 18
Oracle Communications ASAP 7.0.2
Oracle WebLogic Server 10.3.3.0

Benchmark Description

Oracle Communications Service Activation orchestrates the activation of complex services in a flow-through manner across multiple technology domains for both wireline and wireless service providers. This Activation product has two engines: ASAP (Automatic Service Activation Program) and IPSA (IP Service Activator). ASAP covers multiple technologies and vendors, while IPSA focuses on IP-based services.

ASAP converts order activation requests (also referred to as CSDLs) into specific atomic actions for network elements (ASDLs). ASAP performance is measured in throughput and can be expressed either as number of input business orders processed (orders/hour or CSDLs/hour) or as number of actions on network elements (ASDLs/sec). The ratio of CSDL to ASDL depends on the specific telco operator. This workload uses a 1:7 ratio (commonly used by wireless providers), which means that every order translates into actions for 7 network elements. For this benchmark, ASAP was configured to use one NEP (Network Element Processor) per network element.

Key Points and Best Practices

The application and database tiers were hosted on same Netra SPARC T3-1 server.

ASAP has three main components: WebLogic, SARM, NEP. WebLogic is used to receive and translate orders coming in as JMS messages. SARM and NEP, both native applications, perform the core activations functions.

A single ASAP instance delivered slightly under 300k orders/hour, with 27% system utilization. To take better advantage of the SPARC T3 processor's threads, two more instances of ASAP were deployed, reaching 570k orders/hour. The observed ratio between ASAP and Oracle database processor load was 1 to 1.

The Sun Storage 7410 data volumes were mounted via NFS and accessed through the onboard GbE NIC.

A second test was conducted with a more complex configuration of 24 NEPs instead of 7. This simulates the requirements of one of the largest ASAP customers. For this scenario, a single ASAP instances delivered 200k orders/hour.

See Also

Disclosure Statement

Copyright 2011, 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 3/22/2011.

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.

Monday Oct 11, 2010

Sun SPARC Enterprise M9000 Server Delivers World Record Non-Clustered TPC-H @3000GB Performance

Oracle's Sun SPARC Enterprise M9000 server delivered a single-system TPC-H 3000GB world record performance. The Sun SPARC Enterprise M9000 server, running Oracle Database 11g Release 2 on the Oracle Solaris operating system proves the power of Oracle's integrated solution.

  • Oracle beats IBM Power with better performance and price/performance (3 Year TCO). This shows that Oracle's focus on integrated system design provides more customer value than IBM's focus on "per core performance"!

  • The Sun SPARC Enterprise M9000 server is 27% faster than the IBM Power 595.

  • The Sun SPARC Enterprise M9000 server is 22% faster than the HP ProLiant DL980 G7.

  • The Sun SPARC Enterprise M9000 server is 26% lower than the IBM Power 595 for price/performance.

  • The Sun SPARC Enterprise M9000 server is 2.7 times faster than the IBM Power 595 for data loading.

  • The Sun SPARC Enterprise M9000 server is 2.3 times faster than the HP ProLiant DL980 for data loading.

  • The Sun SPARC Enterprise M9000 server is 2.6 times faster than the IBM p595 for Refresh Function.

  • The Sun SPARC Enterprise M9000 server is 3 times faster than the HP ProLiant DL980 for Refresh Function.

  • Oracle used Storage Redundancy Level 3 as defined by the TPC-H 2.12.0 specification, which is the highest level. IBM is the only other vendor to secure the storage to this level.

  • One should focus on the performance of the complete hardware and software stack since server implementation details such as the number of cores or the number of threads will obscure the important metrics of delivered system performance and system price/performance.

  • The Sun SPARC Enterprise M9000 server configured with SPARC VII processors, Sun Storage 6180 arrays, and running Oracle Solaris 10 operating system combined with Oracle Database 11g Release 2 achieved World Record TPC-H performance of 198,907.5 QphH@3000GB for non-clustered systems.

  • The Sun SPARC Enterprise M9000 server is over three times faster than the HP Itanium2 Superdome.

  • The Sun Storage 6180 array configuration (a total of 16 6180 arrays) in this benchmark delivered IO performance of over 21 GB/sec Sequential Read performance as measured by the vdbench tool.

  • This TPC-H result demonstrates that the Sun SPARC Enterprise M9000 server can handle the increasingly large databases required of DSS systems. The server delivered more than 18 GB/sec of real IO throughput as measured by the Oracle Database 11g Release 2 software.

  • Both Oracle and IBM had the same level of hardware discounting as allowed by TPC rules to provide a effective comparison of price/performance.

  • IBM has not shown any delivered I/O performance results for the high-end IBM POWER7 systems. In addition, they have not delivered any commercial benchmarks (TPC-C, TPC-H, etc.) which have heavy I/O demands.

Performance Landscape

TPC-H @3000GB, Non-Clustered Systems

System
CPU type
Memory
Composite
(QphH)
$/perf
($/QphH)
Power
(QppH)
Throughput
(QthH)
Database Available
Sun SPARC Enterprise M9000
2.88GHz SPARC64 VII
512GB
198,907.5 $15.27 182,350.7 216,967.7 Oracle 12/09/10
HP ProLiant DL980 G7
2.27GHz Intel Xeon X7560
512GB
162,601.7 $2.68 185,297.7 142,601.7 SQL Server 10/13/10
IBM Power 595
5.0GHz POWER6
512GB
156,537.3 $20.60 142,790.7 171,607.4 Sybase 11/24/09
Unisys ES7000 7600R
2.6GHz Intel Xeon
1024GB
102,778.2 $21.05 120,254.8 87,841.4 SQL Server 05/06/10
HP Integrity Superdome
1.6GHz Intel Itanium
256GB
60,359.3 $32.60 80,838.3 45,068.3 SQL Server 05/21/07

QphH = the Composite Metric (bigger is better)
$/QphH = the Price/Performance metric (smaller is better)
QppH = the Power Numerical Quantity
QthH = the Throughput Numerical Quantity

Complete benchmark results found at the TPC benchmark website http://www.tpc.org.

Configuration Summary and Results

Server:

Sun SPARC Enterprise M9000
32 x SPARC VII 2.88 GHz processors
512 GB memory
4 x internal SAS (4 x 300 GB)

External Storage:

16 x Sun Storage 6180 arrays (16x 16 x 300 GB)

Software:

Operating System: Oracle Solaris 10 10/09
Database: Oracle Database 11g Release 2 Enterprise Edition

Audited Results:

Database Size: 3000 GB (Scale Factor 3000)
TPC-H Composite: 198,907.5 QphH@3000GB
Price/performance: $15.27/QphH@3000GB
Available: 12/09/2010
Total 3 year Cost: $3,037,900
TPC-H Power: 182,350.7
TPC-H Throughput: 216,967.7
Database Load Time: 3:40:11

Benchmark Description

The TPC-H benchmark is a performance benchmark established by the Transaction Processing Council (TPC) to demonstrate Data Warehousing/Decision Support Systems (DSS). TPC-H measurements are produced for customers to evaluate the performance of various DSS systems. These queries and updates are executed against a standard database under controlled conditions. Performance projections and comparisons between different TPC-H Database sizes (100GB, 300GB, 1000GB, 3000GB and 10000GB) are not allowed by the TPC.

TPC-H is a data warehousing-oriented, non-industry-specific benchmark that consists of a large number of complex queries typical of decision support applications. It also includes some insert and delete activity that is intended to simulate loading and purging data from a warehouse. TPC-H measures the combined performance of a particular database manager on a specific computer system.

The main performance metric reported by TPC-H is called the TPC-H Composite Query-per-Hour Performance Metric (QphH@SF, where SF is the number of GB of raw data, referred to as the scale factor). QphH@SF is intended to summarize the ability of the system to process queries in both single and multi user modes. The benchmark requires reporting of price/performance, which is the ratio of QphH to total HW/SW cost plus 3 years maintenance.

Key Points and Best Practices

  • The Sun Storage 6180 array showed good scalability and these sixteen 6180 arrays showed over 21 GB/sec Sequential Read performance as measured by the vdbench tool.
  • Oracle Solaris 10 10/09 required little system tuning.
  • The optimal 6180 configuration for the benchmark was to set up 1 disk per volume instead of multiple disks per volume and let Oracle Solaris Volume Manager (SVM) mirror. Presenting as many volumes as possible to Oracle database gave the highest scan rate.

  • The storage was managed by SVM with 1MB stripe size to match with Oracle's database IO size. The default 16K stripe size is just too small for this DSS benchmark.

  • All the Oracle files, except TEMP tablespace, were mirrored under SVM. Eight 6180 arrays (128 disks) were mirrored to another 8 6180 arrays using 128-way stripe. IO performance was good and balanced across all the disks with a round robin order. Read performance was the same with mirror or without mirror. With the SVM mirror the benchmark passed the ACID (Atomicity, Consistency, Isolation and Durablity) test.

  • Oracle tables were 128-way partitioned and parallel degree for each table was set to 128 because the system had 128 cores. This setting worked the best for performance.

  • CPU usage during the Power run was not so high. This is because parallel degree was set to 128 for the tables and indexes so it utilized 128 vcpus for the most of the queries but the system had 256 vcpus.

See Also

Disclosure Statement

Sun SPARC Enterprise M9000 198,907.5 QphH@3000GB, $15.27/QphH@3000GB, avail 12/09/10, IBM Power 595 QphH@3000GB, 156,537.3 QphH@3000GB, $20.60/QphH@3000GB, avail 11/24/09, HP Integrity Superdome 60,359.3 QphH@3000GB, $32.60/QphH@3000GB avail 06/18/07, TPC-H, QphH, $/QphH tm of Transaction Processing Performance Council (TPC). More info www.tpc.org.

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.

Monday Aug 23, 2010

Repriced: SPC-1 Sun Storage 6180 Array (8Gb) 1.9x Better Than IBM DS5020 in Price-Performance

Results are presented on Oracle's Sun Storage 6180 array with 8Gb connectivity for the SPC-1 benchmark.
  • The Sun Storage 6180 array is more than 1.9 times better in price-performance compared to the IBM DS5020 system as measured by the SPC-1 benchmark.

  • The Sun Storage 6180 array delivers 50% more SPC-1 IOPS than the previous generation Sun Storage 6140 array and IBM DS4700 on the SPC-1 benchmark.

  • The Sun Storage 6180 array is more than 3.1 times better in price-performance compared to the NetApp FAS3040 system as measured by the SPC-1 benchmark.

  • The Sun Storage 6180 array betters the Hitachi 2100 system by 34% in price-performance on the SPC-1 benchmark.

  • The Sun Storage 6180 array has 16% better IOPS/disk drive performance than the Hitachi 2100 on the SPC-1 benchmark.

Performance Landscape

Select results for the SPC-1 benchmark comparing competitive systems (ordered by performance), data as of August 6th, 2010 from the Storage Performance Council website.

Sponsor System SPC-1 IOPS $/SPC-1
IOPS
ASU
Capacity
(GB)
TSC Price Data
Protection
Level
Results
Identifier
Hitachi HDS 2100 31,498.58 $5.85 3,967.500 $187,321 Mirroring A00076
NetApp FAS3040 30,992.39 $13.58 12,586.586 $420,800 RAID6 A00062
Oracle SS6180 (8Gb) 26,090.03 $4.37 5,145.060 $114,042 Mirroring A00084
IBM DS5020 (8Gb) 26,090.03 $8.46 5,145.060 $220,778 Mirroring A00081
Fujitsu DX80 19,492.86 $3.45 5,355.400 $67,296 Mirroring A00082
Oracle STK6140 (4Gb) 17,395.53 $4.93 1,963.269 $85,823 Mirroring A00048
IBM DS4700 (4Gb) 17,195.84 $11.67 1,963.270 $200,666 Mirroring A00046

SPC-1 IOPS = the Performance Metric
$/SPC-1 IOPS = the Price-Performance Metric
ASU Capacity = the Capacity Metric
Data Protection = Data Protection Metric
TSC Price = Total Cost of Ownership Metric
Results Identifier = A unique identification of the result Metric

Complete SPC-1 benchmark results may be found at http://www.storageperformance.org.

Results and Configuration Summary

Storage Configuration:

80 x 146.8GB 15K RPM drives
4 x Qlogic QLE 2560 HBA

Server Configuration:

IBM system x3850 M2

Software Configuration:

MS Windows 2003 Server SP2
SPC-1 benchmark kit

Benchmark Description

SPC Benchmark-1 (SPC-1): is the first industry standard storage benchmark and is the most comprehensive performance analysis environment ever constructed for storage subsystems. The I/O workload in SPC-1 is characterized by predominately random I/O operations as typified by multi-user OLTP, database, and email servers environments. SPC-1 uses a highly efficient multi-threaded workload generator to thoroughly analyze direct attach or network storage subsystems. The SPC-1 benchmark enables companies to rapidly produce valid performance and price-performance results using a variety of host platforms and storage network topologies.

SPC1 is built to:

  • Provide a level playing field for test sponsors.
  • Produce results that are powerful and yet simple to use.
  • Provide value for engineers as well as IT consumers and solution integrators.
  • Is easy to run, easy to audit/verify, and easy to use to report official results.

Key Points and Best Practices

See Also

Disclosure Statement

SPC-1, SPC-1 IOPS, $/SPC-1 IOPS reg tm of Storage Performance Council (SPC). More info www.storageperformance.org, results as of 8/6/2010. Sun Storage 6180 array 26,090.03 SPC-1 IOPS, ASU Capacity 5,145.060GB, $/SPC-1 IOPS $4.37, Data Protection Mirroring, Cost $114,042, Ident. A00084.

Repriced: SPC-2 (RAID 5 & 6 Results) Sun Storage 6180 Array (8Gb) Outperforms IBM DS5020 by up to 64% in Price-Performance

Results are presented on Oracle's Sun Storage 6180 array with 8 Gb connectivity for the SPC-2 benchmark using RAID 5 and RAID 6.
  • The Sun Storage 6180 array outperforms the IBM DS5020 system by 62% in price-performance for SPC-2 benchmark using RAID 5 data protection.

  • The Sun Storage 6180 array outperforms the IBM DS5020 system by 64% in price-performance for SPC-2 benchmark using RAID 6 data protection.

  • The Sun Storage 6180 array is over 50% faster than the previous generation systems, the Sun Storage 6140 array and IBM DS4700, on the SPC-2 benchmark using RAID 5 data protection.

Performance Landscape

Select results from Oracle and IBM competitive systems for the SPC-2 benchmark (in performance order), data as of August 7th, 2010 from the Storage Performance Council website.

Sponsor System SPC-2 MBPS $/SPC-2 MBPS ASU Capacity (GB) TSC Price Data
Protection
Level
Results Identifier
Oracle SS6180 1,286.74 $56.88 3,504.693 $73,190 RAID 6 B00044
IBM DS5020 1,286.74 $93.26 3,504.693 $120,002 RAID 6 B00042
Oracle SS6180 1,244.89 $50.40 3,504.693 $62,747 RAID 5 B00043
IBM DS5020 1,244.89 $81.73 3,504.693 $101,742 RAID 5 B00041
IBM DS4700 823.62 $106.73 1,748.874 $87,903 RAID 5 B00028
Oracle ST6140 790.67 $67.82 1,675.037 $53,622 RAID 5 B00017
Oracle ST2540 735.62 $37.32 2,177.548 $27,451 RAID 5 B00021
Oracle ST2530 672.05 $26.15 1,451.699 $17,572 RAID 5 B00026

SPC-2 MBPS = the Performance Metric
$/SPC-2 MBPS = the Price-Performance Metric
ASU Capacity = the Capacity Metric
Data Protection = Data Protection Metric
TSC Price = Total Cost of Ownership Metric
Results Identifier = A unique identification of the result Metric

Complete SPC-2 benchmark results may be found at http://www.storageperformance.org.

Results and Configuration Summary

Storage Configuration:

Sun Storage 6180 array with 4GB cache
30 x 146.8GB 15K RPM drives (for RAID 5)
36 x 146.8GB 15K RPM drives (for RAID 6)
4 x PCIe 8 Gb single port HBA

Server Configuration:

IBM system x3850 M2

Software Configuration:

Microsoft Windows 2003 Server SP2
SPC-2 benchmark kit

Benchmark Description

The SPC Benchmark-2™ (SPC-2) is a series of related benchmark performance tests that simulate the sequential component of demands placed upon on-line, non-volatile storage in server class computer systems. SPC-2 provides measurements in support of real world environments characterized by:
  • Large numbers of concurrent sequential transfers.
  • Demanding data rate requirements, including requirements for real time processing.
  • Diverse application techniques for sequential processing.
  • Substantial storage capacity requirements.
  • Data persistence requirements to ensure preservation of data without corruption or loss.

Key Points and Best Practices

  • This benchmark was performed using RAID 5 and RAID 6 protection.
  • The controller stripe size was set to 512k.
  • No volume manager was used.

See Also

Disclosure Statement

SPC-2, SPC-2 MBPS, $/SPC-2 MBPS are regular trademarks of Storage Performance Council (SPC). More info www.storageperformance.org, results as of 8/9/2010. Sun Storage 6180 Array 1,286.74 SPC-2 MBPS, $/SPC-2 MBPS $56.88, ASU Capacity 3,504.693 GB, Protect RAID 6, Cost $73,190, Ident. B00044. Sun Storage 6180 Array 1,244.89 SPC-2 MBPS, $/SPC-2 MBPS $50.40, ASU Capacity 3,504.693 GB, Protect RAID 5, Cost $62,747, Ident. B00043.

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
« April 2014
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
30
   
       
Today