Java Business Benchmark

Description

OS ENVIRONMENT
Java Business Benchmark
Performance on the Dell PowerEdge 6650 Server
™Java applications are becoming increasingly popular in enterprise environments, neces-
sitating the measurement of Java performance on servers. This article discusses SPECjbb2000—
an industry-standard benchmark that measures Java server performance—and examines
™ ™ ™results on the Intel Xeon processor-based Dell PowerEdge 6650 server.
BY RAMESH RADHAKRISHNAN, PH.D.
he Standard Performance Evaluation Corporation the Java objects in memory, no disk I/O occurs in
® ™T(SPEC ) introduced the SPECjbb 2000 benchmark in SPECjbb2000. No network I/O is generated or measured
June 2000 to evaluate the performance of servers running in SPECjbb2000, because users do not reside on exter-
™typical Java business applications. The benchmark can be nal client systems.
®run on different operating systems, including the Microsoft One active customer is assigned to each warehouse that
® ® ®Windows , Linux , and UNIX operating systems. is initialized when running the benchmark. A warehouse
SPECjbb2000 simulates a wholesale supplier, with is a unit of approximately 25 MB of data, stored in Btrees.
multiple warehouses that serve several districts. The Warehouses map directly to Java threads, resulting in a
benchmark uses a three-tier client-server scheme to multithreaded workload that scales well on multiple proces-
model the company, emphasizing the middle tier (see sors. The number of threads increases as ...

Sujets

• Java Business Integration
• Dell Diamond
• Server Message Block
• Système d'exploitation
• Spectres
• Tiergarten

Informations

• Publié par: Shuim
• Nombre de lectures: 69
• Langue: English

Extrait

OS ENVIRONMENT

Java Business Benchmark Performance on the Dell PowerEdge 6650 Server

™ Java applicationsare becoming increasingly popular in enterprise environments, neces- sitating the measurement of Java performance on servers. This article discusses SPECjbb2000— an industry-standard benchmark that measures Java server performance—and examines * ™™ ™ results on the Intel Xeon processor-basedDell PowerEdge 6650server.

BY RAMESH RADHAKRISHNAN, PH.D. he Standard Performance Evaluation Corporation T ® ™ (SPEC ) introducedthe SPECjbb2000 benchmark in June 2000 to evaluate the performance of servers running ™ typical Javabusiness applications. The benchmark can be ® run on different operating systems, including the Microsoft ® ®® Windows , Linux , and UNIXoperating systems. SPECjbb2000 simulates a wholesale supplier, with multiple warehouses that serve several districts. The benchmark uses a three-tier client-server scheme to model the company, emphasizing the middle tier (see Figure 1). The tiers mimic a typical order-processing application: Users in the first tier generate inputs that result in the execution of business logic in the second tier. In turn, the second tier accesses a database on the third tier to satisfy customer requests. Customers can per- form operations such as placing new orders or deter- mining or viewing the status of an existing order. Other operations such as processing orders for delivery, enter- ing customer payments, and checking stock levels are generated within the company. SPECjbb2000 is loosely based on the TPC-C bench- mark from the Transaction Processing Performance Council (TPC) and borrows from the TPC-C specifica- tion for its schema, input generation, and operation pro- file. SPECjbb2000 replaces database tables with Java classes and replaces data records with Java objects. Because binary trees (Btrees) or other data objects hold

www.dell.com/powersolutions

the Java objects in memory, no disk I/O occurs in SPECjbb2000. No network I/O is generated or measured in SPECjbb2000, because users do not reside on exter- nal client systems. One active customer is assigned to each warehouse that is initialized when running the benchmark. Awarehouse is a unit of approximately 25 MB of data, stored in Btrees. Warehouses map directly to Java threads, resulting in a multithreaded workload that scales well on multiple proces- sors. The number of threads increases as the number of warehouses scales up during the benchmark run. Although SPECjbb2000 is not a complete online transaction processing (OLTP) benchmark like TPC-C, it is a good approximation of a large Java business application. The benchmark tests the performance of software components such as the Java Virtual Machine ™ (JVM ),just-in-time (JIT) compiler, garbage collection, and some aspects of the operating system. It also mea- sures performance of CPUs, memory hierarchy, and the scalability of shared-memory processors on the hardware side.

Measuring performance using SPECjbb2000 SPECjbb2000 calculates system throughput as the number of business operations that are performed per second, and it uses this rate to compare the performance of different servers. The metric is a composite number calculated by

POWER SOLUTION 115

OS ENVIRONMENT

averaging throughput obtained at different points of the run. SPECjbb2000 determines the metric as follows:

• The number of warehouses is increased from one to at least twice the number of warehouses that is expected to produce peak throughput. The peak typically occurs when the number of warehouses equals the number of CPUs in the system. To be considered a valid benchmark run, the test must calculate every point from one to eight warehouses. The user can specify how many warehouses to include in a • benchmark run; this number typically is based on the number of processors in the system. This article assumes that the peak occurs atnwarehouses. • Throughput for all the points—fromnwarehouses to 2n warehouses—is averaged to calculate the SPECjbb2000 metric. To achieve a valid result, the user should set the upper limit on the number of warehouses to be at least 2n, wherenis the number of processors in the system or warehouse when maximum throughput is achieved— whichever is higher.

Business logic engine

 Operations Company Systemper secondJava Virtual MachinePublished   Dell PowerEdge6650 76136BEA WebLogic JRockit 8.0January 2003  (Windows2000) Dell PowerEdge6650 73532WebLogic JRockit 8.0 (Linux)January 2003  IBM IBMeServer 73319Java2 Runtime Environment (J2RE)December 2002 ™  xSeries360 1.4.0IBM Windows 32 build  cn140-20020902(JIT enabled: jitc)  ™ Hewlett- HPServer 63414Java HotSpotVM v. 1.3.1.08 onOctober 2002   Packard rx5670HP-UX 11iv. 1.6 for Intel Itanium2 Dell PowerEdge6650 57245WebLogic JRockit 7.0October 2002

-

 Operations Company SystemCPUs persecond JavaVirtual Machine  ™ IBM IBMRS/6000 872437 J2RE1.3.0 IBM AIX 5Lfor PowerPC  ModelM80 (64-bitJVM) ™ IBM IBMeServer pSeries8 72437J2RE 1.3.0 IBM AIX 5L for PowerPC  660Model 6M1(64-bit JVM) IBM IBMeServer pSeries12 71303J2RE 1.3.0 IBM build (JIT enabled: jitc)  680 ™ IBM IBMeServer iSeries12 69841iSeries Developer Kit for Java 1.3.0  840-2461 HP HPServer rp74008 63336Java HotSpot VM v. 1.3.1.02

1 SPECjbb2000 benchmark results as of February 24, 2003. For the latest reports, visit http://www.spec.org/jbb2000/results/jbb2000.html.

- -