Outstanding PRIMEPOWER Performance

Tests d'évaluation


		Benchmarks
		Outstanding PRIMEPOWER performance
		PRIMEPOWER systems are highly scalable UNIX servers from Fujitsu Siemens Computers. The product lineup includes a solution for every requirement, from small and mid-sized workgroup servers and compute nodes to enterprise models for mission critical applications. These industry-grade systems offer unmatched scalability. To document their performance, Fujitsu Siemens has published results for several leading industry standard benchmarks, which are described below. SPECjAppServer2002 benchmark Measures the performance and price-performance of Java Enterprise Application Servers in a complete end-to-end Web application. The SPECjAppServer2002 workload is modeled after a manufacturing, supply-chain management and order/inventory environment. Performance results are influenced primarily by hardware configuration, J2EE application software and database software. The two performance metrics (TOPS and $/TOPS) are determined by the number of order transactions plus the number of manufacturing work orders per second. The results are classified into four categories, depending on how the application and database servers are distributed on the systems being tested. See SPECjAppServer2002 for a more detailed description. Oracle Applications Benchmark Makes a result comparison possible by using Oracle’s standard product software, a standardized module and workload selection. Several modules from Oracle Financial Applications like Accounts Payable (AP), General Ledger (GL), Purchase Orders (PO), Inventory (INV), and others are included. The benchmark, which determines the number of supported users, is measured in a three-tier client/server environment (Remote Terminal Emulators, Concurrent Processing and Application Servers, Database Server) where the limiting resource is the database server. See Oracle Applications Benchmark for a more detailed description. SAP SD Standard Application Benchmark SAP has defined a variety of benchmarks accompanying its R/3 product releases since 1993, growing out of sizing considerations for SAP R/3 users. The most popular one is the Sales and Distribution (SD) benchmark. It represents a supply chain business scenario where a customer order is placed, delivery of goods is scheduled, and an invoice is written. The benchmark determines the number of users that can be supported. It can be measured in two versions: “two-tier” with application software and database on the same system; or “three-tier” with application software and database on different systems. See http://www.sap.com/benchmark for a more detailed description of the SAP Standard Application Benchmarks. SAP ATO Standard Application Benchmark The Assembly To Order (ATO) benchmark is similar to SAP’s older SD benchmark (see above), but has more dialogue steps per order and an increased intrinsic complexity. The benchmark determines the number of users that can be supported and can be measured in two versions: “two-tier” with application software and database on the same system; or “three-tier” with application software and database on different systems. See http://www.sap.com/benchmark for a more detailed description. SAP Business Information Warehouse Standard Application Benchmark As SAP Business Information Warehouse is a important component of SAP Netweaver, the SAP BW benchmark has become the most popular benchmark besides the SD benchmark. The SAP Business Information Warehouse (BW) Benchmark consists of two phases, a load phase and a query phase. It can be measured in two versions: “two-tier” with application software and database on the same system; or “three-tier” with application software and database on different systems. See http://www.sap.com/benchmark for a more detailed description. SPEC CPU2000 benchmarks SPEC CPU benchmarks are collections of CPU-intensive application or system/utility programs. They show the performance of the CPU, the memory system (including caches), and the compiler’s ability to generate optimal code for the system being tested. The integer collection (metric SPECint...) consists of 12 integer programs representative of commercial or system codes. The FP collection (metric SPECfp...), meanwhile, consists of 14 numeric-scientific programs. Measurements are possible for “speed” with one CPU active, or for throughput/”SPECrate” with several CPUs executing identical copies of the benchmark. See SPEC CPU2000 for a more detailed description. SPEC OMP2001 benchmarks SPEC OMP benchmarks measure the execution speed of a suite of 11 numeric-scientific programs using the OpenMP model of parallel execution of a single program on a multi-CPU system. The individual benchmarks are almost identical to those in SPEC's CPU2000 FP collection. The run rules allow the insertion of OpenMP directives at certain source code locations, as customary in parallel computing. Measurements are possible for a “medium” (M) and a “large” (L) input data set. See SPEC OMP2001 for a more detailed description. SPECweb99 benchmark SPEC’s web server benchmark measures the performance of an Internet or Intranet Web server (hardware and software) responding to HTTP requests generated by several client systems. SPECweb99 determines how many simultaneous connections can be maintained with the benchmark-defined load, consisting of static as well as dynamic HTTP requests. File-size distributions are derived from real-world Web server statistics. See SPECweb99 for a more detailed description. SPECjbb2000 benchmark SPEC’s industry standard benchmark for Java server computing models the middle tier of three-tier Java server applications. It measures the hardware (CPU, caches, memory) as well as the software (Operating System, Java Virtual Machine) of the system being tested. SPECjbb2000 is loosely modeled after the TPC-C benchmark, where customers initiate a set of operations such as placing new orders or requesting the status of a given order. Because of the specific implementation, however, the results cannot be compared with TPC-C results. See SPECjbb2000 for a more detailed description. TOP500 / LINPACK benchmark The popular TOP500 list is compiled twice a year, listing large installations of supercomputers ranked according to their performance on the LINPACK benchmark. LINPACK evolved from a core algorithm in a popular linear equations package. In the version employed for the TOP500 list, the use of high-performance libraries and parallel execution on multiple CPUs and multiple nodes in a cluster is permitted, since it is typical for high-performance numeric computing. See TOP500 for a more detailed benchmark description. TPC-C benchmark This Transaction Processing Performance Council (TPC) benchmark measures OLTP performance in a three-tier client/server setting where many virtual users concurrently execute transactions against a single database. The environment models a wholesale supplier with five types of transactions, such as placing a new order and requesting the status of a given order. The benchmark is very update-intensive, with a high number of disk I/O requests. Benchmark results are quoted in transactions per minute (tpmC), together with the price per tpmC. See TPC-C for a more detailed benchmark description. TPC-H benchmark This Transaction Processing Performance Council (TPC) benchmark measures decision support performance. TPC-H represents a workload for decision support systems that examines large volumes of data, executes queries with a high degree of complexity, and gives answers to critical business questions. Results are quoted as Composite Query-per-Hour Performance (QphH@Size), together with the price per QphH@size. Five different database sizes (100, 300, 1000, 3000 & 10000 GB) are defined for the benchmark, although result comparisons are permitted only within a specific size class. See TPC-H for a more detailed benchmark description.

Benchmarks