IBM. IBM ^ pseries, IBM RS/6000 and IBM NUMA-Q Performance Report

Transcription

1 IBM IBM ^ pseries, IBM RS/6000 and IBM NUMA-Q Performance Report October 7,

2 Table of Contents PERFORMANCE of IBM WEB SERVER SYSTEMS... 3 Section - and LINPACK PERFORMANCE... RS/6000 SP s... 4 Section a - 95 PERFORMANCE... 4 RS/6000 SP s... 5 Section 2 - MULTI-USER PERFORMANCE... 5 RS/6000 SP s... 6 Section 2a - MULTI-USER 95 AND web96 PERFORMANCE... 8 RS/6000 SP s... 9 Section 3 - TPC-C PUBLISHED RESULTS... 0 Section 4 - TPC-H PUBLISHED RESULTS... 0 Section 5 - NOTESBENCH PUBLISHED RESULTS... Section 6 - JAVA BENCHMARKS (jvm98, jbb) PUBLISHED RESULTS.. NOTES ON PERFORMANCE BENCHMARKS AND VALUES

3 PERFORMANCE of IBM WEB SERVER SYSTEMS October 7, This is an update of the October 03, report. The TPC-C and the updated ROLTP results of the IBM ^ pseries 680 are presented in this report. Adjusted ROLTP results of RS/6000-S80 configurations are also included. The adjusted results are based on software levels identical to the pseries 680 ROLTP results. The adjusted S80 ROLTP results enable better comparison with the p680 results, but should not be used to compare any other server. Section One of this report includes the results. replaced the 95 benchmarks. The 95 results are included in Section a to aid the transition. and LINPACK results for all of the current and recently withdrawn RS/6000 systems are presented in this section. Section Two, Multiuser Performance of several systems are presented. The maximum memory on is increased to 6GB. This lifted the memory constraint on the ROLTP results of the. The ROLTP result of the POWER3 SMP Thin Node (67.7) is limited by the number of PCI slots (two). The 95 multiuser numbers and web96 numbers are included in section 2a. Section Three reflects the published TPC-C results to date. Section Four, TPC-H results of an RS/6000 SP system with 32 POWER3 SMP Wide Nodes and the 8, 2 and 6-quad NUMA-Q E40 systems are provided. Section Five reflects the published NotesBench results to date for RS/6000 systems including the NotesBench Version 5 result. Section Six reflects the published jvm98 and jbb Java benchmarks. All performance measurements for the IBM ^ pseries and IBM RS/6000 systems were made with systems running the AIX operating system. For new and upgraded systems, AIX Version 4.3 was used. All other systems used previous versions of AIX. All performance measurements for the NUMA-Q systems were made with the systems running the DYNIX/ptx operating system. 3

4 Section - and LINPACK PERFORMANCE /28 RS64 III/ /28 RS64 III/ RS64 II/ / /28 RS64 III/ / / # / #F / / B / / 80 P3/ #43P P P P P P #43P #43P-40n #43P-40u TPP SP DP LINPACK fp int L RS/6000 SP s / / POWER3 T/W P3/ # / T/W /28 60 P2SC #60 Thin TPP SP DP LINPACK fp int Proc./ L 4

5 Section a - 95 PERFORMANCE /28 RS64 III/ /28 RS64 III/ RS64 II/ / /28 RS64 III/ / / # / #F / / B / / P3/ #43P P P P P P #43P #43P-40n #43P-40u base95 fp95 base95 int95 L RS/6000 SP s / / POWER3T/W P3/ # #POWER3T/W P3/ / T/W /28 60 P2SC #60 Thin base95 fp95 base95 int95 L 5

6 Section 2 - MULTI-USER PERFORMANCE /28 RS64 III/ /28 RS64 III/ 57. RS64 II/ RS64 II/3 3.9 RS64 II/2 6.7 RS64 II/ 32.8 / /3 7.9 /2 0.0 / / /28 RS64 III/ /28 RS64 III/ /28 RS64 III/ 32.8 / /3 7.9 /2 0.0 / /4 # 66 /3 # /2 # / # /2 #F / #F / / / / B / / / / 440 P3/2 #43P P P P P P #43P #43P-40n #43P-40u web99 Rel. OLTP Perf. L 6

7 L Rel. OLTP Perf. web99 #S7A #S7A #S7A S80 S80 S80 S80 *S80 *S80 *S80 *S80 *p680-s85 *p680-s85 *p680-s85 *p680-s85 RS64 III/4 /2 /4 /6 /8 RS64 II/4 RS64 II/8 RS64 II/2 RS64 III/2 RS64 III/4 RS64 III/8 RS64 III/2 RS64 III/8 RS64 III/24 RS64 III/2 RS64 III/8 RS64 III/24 RS64 IV/6 RS64 IV/2 RS64 IV/8 RS64 IV/ /28 28/28 28/28 28/28 28/28 28/28 28/28 28/28 28/28 28/28 28/28 28/28 28/28 28/28 28/28 28/28 28/28 28/ a a a a a -Adjusted S80 ROLTP results (as of October ) are based on software levels identical to the p680 results. The adjusted S80 results enable better comparison with p680, but should not be used to compare any other server. 7

8 RS/6000 SP s / / / POWER3 W 67.7 POWER3 T 4 /2 POWER3T/W 8.3 P3/8 # 6 P3/6 # 43.3 P3/4 # 23.0 P3/2 # P3/2 #POWER3T/W 0.5 P3/ #POWER3T/W /4 T/W /2 T/W /28 60 P2SC #60 Thin web99 Rel. OLTP Perf. L # Nodes Proc./ # CPUs 8

9 Section 2a - MULTI-USER 95 AND web96 PERFORMANCE /28 RS64 III/2 S /28 RS64 III/8 20 RS64 II/2 #S7A / / / / / /28 RS64 III/ /28 RS64 III/ RS64 II/ RS64 II/ RS64 II/ RS64 II/ / / / / / /28 RS64 III/ /28 RS64 III/ / / / / /4 # /2 # / # /2 #F / #F / / P3/2 #43P P3/ #43P-260 web L 9

10 RS/6000 SP s Proc./ # CPUs # Nodes L web96 #60 Thin T/W T/W #POWER3T/W #POWER3T/W # # # # POWER3 T/W POWER3 T POWER3 W POWER3 W POWER3 W POWER3 W P2SC /2 /4 P3/ P3/2 P3/2 P3/4 P3/6 P3/8 /8 /2 / / Section 3 - TPC-C PUBLISHED RESULTS # CPUs # Nodes tpmc $/tmpc Database AIX Availability Date # #S70 #S70 #S7A S80 *p680 /4 /4 /8 RS64/2 RS64 II/2 RS64 II/2 RS64 II/4 RS64 III/24 RS64 III/8 RS64 IV/ , , ,65.3 8, , , , , , , , Sybase.5 Sybase.5 Sybase.5 Oracle V8. Oracle V8. Oracle OPS Oracle V85 Oracle V86 Oracle V86 Oracle V87 Oracle V /09/98 02/20/98 09/30/97 09/02/98 0/2/99 06/28/99 /9/99 03/0/00 06/09/00 09/30/00 04/3/0 0

11 Section 4 - TPC-H PUBLISHED RESULTS TPC-H TB: RS/6000 SP with POWER3 SMP nodes # Proc. Nodes QphH QppH QthH $/ QphH AIX Database Avail. Date 32 2, ,82.3 2, DB2 UDB7. 8/5/00 TPC-H 300GB: NUMA-Q E40 (Avail. Date: 05/25/00): Intel Pentium III Xeon # $/ Proc. Quads QphH QppH QthH QphH ptx Database Avail. Date Xeon/4 Xeon/4 Xeon/ DB2 UDB7. DB2 UDB7. DB2 UDB7. 8/3/00 8/3/00 8/5/00 Section 5 - NOTESBENCH PUBLISHED RESULTS L Users TPM Response Time $/User Version #43P-40n #S70 S80 / / RS64 II/4 RS64 II/2 RS64 III/24 RS64 III/8 28/28 28/ b Section 6 - JAVA BENCHMARKS (jvm98, jbb) PUBLISHED RESULTS L jvm98 (256MB) jbb ops/sec 44P-70 p680-s85 RS64 IV/ / ,644 4,644 56,834 ======================================================================================== * - Values are new or changed from October 3, version. # - These systems have been withdrawn from marketing and are included for historical purposes. P2SC - POWER2 Super Chip P3 - POWER3

12 NOTES ON PERFORMANCE BENCHMARKS AND VALUES The performance benchmarks and the values shown here were derived using particular, well configured, development-level computer systems. Unless otherwise indicated for a system, the values were derived using 32-bit applications and external cache if external cache is supported on the system. All performance benchmark values are provided "AS IS" and no warranties or guarantees are expressed or implied by IBM. Actual system performance may vary and is dependent upon many factors including system hardware configuration and software design and configuration. Buyers should consult other sources of information to evaluate the performance of systems they are considering buying and should consider conducting application oriented testing. For additional information about the performance benchmarks, values and systems tested, please contact your IBM local Branch Office or IBM Authorized Reseller or access the following on the Web: - TPC - LINPACK - Unless otherwise indicated for a system, the performance benchmarks were conducted using AIX v 4.2 or v 4.3. IBM C for AIX v 3.6. and XL Fortran v and v were the compilers used in the benchmark tests. The preprocessors used in the benchmark tests include KAP 3.2 for Fortran and KAP/C.4.2 from Kuck & Associates and VAST-2 v X8 from Pacific-Sierra Research. The preprocessors were purchased sepaly from these vendors. The following and LINPACK benchmarks reflect the performance of the microprocessor, memory architecture and compiler of the tested system: int95 - component-level benchmark that measures integer performance. Result is the geometric mean of eight tests that comprise the CINT95 benchmark suite. All of these are written in C language. base95 is the result of the same tests in CINT95 with a maximum of four compiler flags that must be used in all eight tests Geometric average of the eight s from the integer tests (CINT95). 95 is the result of the same tests as CINT95 with a maximum of four compiler flags that must be used in all eight tests. fp95 - component-level benchmark that measures floating-point performance. Result is the geometric mean of ten tests, all written in FORTRAN, that are included in the CFP95 benchmark suite. base95 is the result of the same tests in CFP95 with a maximum of four compiler flags that must be used in all ten tests Geometric average of the ten s from floating-point tests (CFP95). _95 is the result of the same tests as CFP95 with a maximum of four compiler flags that must be used in all ten tests. int - New component-level benchmark that measures integer performance. Result is the geometric mean of twelve tests that comprise the CINT benchmark suite. All of these are written in C language except for one which is in C++. base is the result of the same tests in CINT with a maximum of four compiler options that must be used in all twelve tests. - Geometric average of the twelve s from the integer tests (CINT). _base is the result of the same tests as CINT with a maximum of four compiler options that must be used in all twelve tests. 2

13 fp - New component-level benchmark that measures floating-point performance. Result is the geometric mean of fourteen tests, all written in FORTRAN and C languages, that are included in the CFP benchmark suite. base is the result of the same tests in CFP with a maximum of four compiler options that must be used in all fourteen tests. - Geometric average of the fourteen s from floating-point tests (CFP). _base is the result of the same tests as CFP with a maximum of four compiler options that must be used in all fourteen tests. web96 - Maximum number of Hypertext Transfer Protocol (HTTP) operations per second achieved on the web96 benchmark without significant degradation of response time. The Web server software is IBM HTTP Server or ZEUS from Zeus Technology Ltd. web99 - Number of conforming, simultaneous connections the Web server can support using a predefined workload. The web99 test harness emulates clients sending the HTTP requests in the workload over slow Internet connections to the Web server. The Web server software is Zeus from Zeus Technology Ltd. jvm99 - The jvm98 benchmark suite contains eight different tests. Each test measures the time it takes to load the program, verify the class files, compile on the fly if a JIT compiler is used, and execute the test. A geometric mean is used to compute the a composite score. Test scores are normalized against a reference machine. Higher scores indicate better performance. jbb - Expressed in operations per second. jbb represents an order processing application for a wholesale supplier. It is implemented as a Java program emulating a three-tier client/server system with emphasis on the middle tier. LINPACK DP (Double Precision) - n=00 data array. Units are Megaflops. LINPACK SP (Single Precision) - n=00 data array. Units are Megaflops. LINPACK TPP (Toward Peak Performance) - n=,000 data array. Units are Megaflops. ESSL Version 3.. or 3..2 was used in this test. The following Transaction Processing Council (TPC) benchmarks reflect the performance of the microprocessor, memory subsystem, disk subsystem and some portions of the network: tpmc - TPC Benchmark C throughput measured as the average number of transactions processed per minute during a valid TPC-C configuration run of at least twenty minutes. $/tpmc - TPC Benchmark C price-performance ratio reflects the estimated five year total cost of ownership for system hardware, software and maintenance and is determined by dividing such estimated total cost by the tpmc for the system. QppH is the power metric of TPC-H and is based on a geometric mean of the 7 TPC-H queries, the insert test and the delete test. It measures the ability of the system to give a single user the best possible response time by harnessing all available resources. QppH is scaled based on database size from 30GB to TB. QthH is the throughput metric of TPC-H and is a classical throughput measure characterizing the ability of the system to support a multiuser workload in a balanced way. A number of query users is chosen, each of which must execute the full set of 7 queries in a different order. In the background, there is an update stream that runs a series of insert/delete operations. QthH is scaled based on the database size from 30GB to TB. QphH is the geometric mean of the power tests (QppH) and the throughput tests (QthH). 3

14 $/QphH is the price/performance metric for the TPC-H benchmark where QphH is the geometric mean of QppH and QthH. The price is the five year cost of ownership for the tested configuration and includes maintenance and software support. NotesBench is the driver program to test various aspects of Lotus Notes. It is designed to execute the commands in customized workload scripts, simulating Notes client actions. Notesbench tests include "mail only", "mail and database". All IBM results have been published with the "mail only" workload. Total Users - number of active users supported in the workload, each producing approximately one transaction/minute. TPM - transactions per minute (NotesMark) Average Response Time - average time for a transaction to be completed for an average user action. $/User - Total cost of the hardware and software including discounts quoted by a supplier. NOTES ON PERFORMANCE ESTIMATES Relative OLTP (ROLTP) is an estimate of commercial processing performance derived from an IBM analytical model. The model simulates some of the system's operations such as CPU, cache and memory. However, the model does not simulate disk or network I/O operations. Although general database and operating system parameters are used, the model does not reflect specific databases or AIX version or releases. Unless otherwise indicated for a system, the model assumes the use of 32-bit applications. Unless otherwise indicated for a system, ROLTP is estimated only at the time the system is introduced. An IBM RS/ is the baseline reference system and has a value of.0. Although ROLTP may be used to compare estimated commercial processing performance, actual system performance may vary and is dependent upon many factors including system hardware configuration and software design and configuration. All performance estimates are provided "AS IS" and no warranties or guarantees are expressed or implied by IBM. Buyers should consult other sources of information, including system benchmarks, to evaluate the performance of a system they are considering buying. 4

15 IBM Copyright IBM Corporation IBM Corporation Marketing Communications Sever Group Route 00 Somers, New York 0589 Printed in the United States of America 0-00 All Rights Reserved This publication was developed for products and/or services offered in the United States. IBM may not offer the products, features, or services discussed in this publication in other countries. The information may be subject to change without notice. Consult your local IBM business contact for information on the products, features and services available in your area. This equipment is subject to FCC rules. It will comply with the appropriate FCC rules before final delivery to the buyer. IBM hardware products are manufactured from new parts, or new and used parts. Regardless, our warranty terms apply. Information concerning non-ibm products was obtained from the suppliers of these products. Questions on the capabilities of the non-ibm products should be addressed with the suppliers. IBM, the IBM logo, the e-business logo, AIX, DYNIX/ptx, pseries, RS/6000 and SP are trademarks or registered trademarks of International Business Machines Corporation in the United States or other countries or both. Pentium is a registered trademark of Intel Corporation in the United States, other countries, or both. Other company, product, and service names may be trademarks or service marks of others. The IBM home page on the Internet can be found at 5