ANNOUNCEMENT: THE MOST CURRENT RESULTS OF NFS BENCHMARKING ON THE SERIES 800 *********************************************************************** We have great news for our series 800 business servers. Our current benchmark testing indicates that we are the market leader for general purpose servers in NFS operational performance. NFS benchmarks indicate how well an NFS server will respond to continuous operation requests from NFS clients. The benchmark used in these test below is the current industry accepted NFS benchmark, NHFSstones version 1.22. In comparison with our competition, only Auspex today is publishing numbers ahead of HP. Auspex NFS dedicated servers provide around 1400 operations per second. However, an Auspex server is dedicated only for NFS traffic. HP series 800 business servers can be used as high-performance NFS servers, compute servers, database servers or other purposes. THE PERFORMANCE RESULTS: **** NFS (Network File System) Performance **** **** Nhfsstone 1.22 **** **** **** **** 9000/877 - 160 MB - 4 LAN - 8 FL Disks w/ChanSpan - 8.02 **** **** Asynchronous NFS mounts **** **** **** ****** -->>>>> 1000.33 IOPS @ 47.43 ms <<<<<--- ****** ----------------------- GSY Performance would like to announce the achievement of a major milestone in our NFS performance efforts: 1000+ NFS (Using Nhfsstone version 1.22) IOPS (I/O operations per seconds) on the 877. This was achieved on released series 800 hardware and software using 160 MB memory, 8 Fiberlink disks and 4 LAN cards (on 4 LANs), and is the first known 1000+ result for any general purpose UNIX system. In similar NHFSStone 1.22 testing, the following series 800 systems acheived these results: Model 867 with 3 LANS: 746 IOPS @ 49ms with SYNCHRONOUS MOUNTS 868 IOPS @ 41ms with ASYNCHROUNOUS MOUNTS This data is in the current Series 800 Performance Brief WHAT IS AN NHFSSTONES 1.22 BENCHMARK? NHFSSTONE 1.2.2 uses several client or driver systems to present a specific NFS workload to a server system under test. The response time for the server to each request is measured and then the average response time is reported in milli-seconds along with the throughput in Input/Output operations per second (IOPS). Please note this measure only measures server performance. Client issues will also need to be considered. The overall direction in NFS benchmarks will continue to be focused on the server. You may have heard of other NFS benchmark numbers as well. Today, NHFSSTONES version 1.22 is the publically available, defacto standard for measuring NFS performance. A new measure called LADDIS is being developed and has been proposed to SPEC for acceptance and become the official NFS benchmark standard. We expect it to be adopted in the fall of 1992 and will publish LADDIS results after adoption. WHAT OTHER BENCHMARKS ARE BEING USED? When you compare NFS benchmark results, please be aware that several versions of NFS benchmarks are being used by vendors. As a summary, these include: 1. NFSSTONE--This benchmark was developed by Encore and Software Tool & Die. It was presented at USENIX in the summer of '89 and has a completely different workload than any of the NHFSSTONE benchmarks. It is not being used by vendors today. 2. NHFSSTONE 1.22--This benchmark was developed by Legato Systems and distributed in 1990. It is publically available and is used extensively for NFS benchmarking. 3. NHFSSTONE 2.03--An early version of 2.04 used by SUN. It is not available for any other vendor's use. 4. NHFSSTONE 2.04--This benchmark is a modification of 1.2.2 used by Auspex and Data General to avoid the dependency on the client file system and NFS code. It is not publically available. 5. Pre-LADDIS--This is an interim version of a replacement for NHFSSTONE called LADDIS which has been submitted to SPEC for acceptance as a standard benchmark. HP will not publish Pre-LADDIS numbers as there are numerous different versions. LADDIS numbers will be available once LADDIS is accepted as a NFS/System File server benchmark. BE AWARE WHEN YOU COMPARE: The various versions of NHFSSTONE are not comparable. This is especially true with version 1.22 and 2.03 {SUN's version}. Version 2.03 is known to generate less work for a given IOPS value. This can skew the results. SERIES 800 PERFORMANCE: No other general purpose UNIX system offers this level of NFS performance. in fact, only one other vendor offers performance above 1000 IOPS. This vendor is AUSPEX which sells special purpose dedicated NFS servers. HP's series 800 business server product line offers excellent NFS server performance and is also an industry leading database and OLTP server. BENCHMARK DETAIL: The following graph depicts the detailed results of the test: Please note that the delay is well under the industry acceptable delay of maximum 70 ms. HP subjects our systems to the rigorous requirement of no more delay than 50 ms. Nhfsstone 1.22 on 877 ............................................................ . .. 45 . . . . D . . e 40 . . l . . a . . . y 35 . . . . . . i . . . n 30 . . .. . . .. m . .. . s 25 . .. .. . .... .... . ..... ..... 20 . ............... ............................................................. 200 400 600 800 1000 IOPS 4 LAN, 8 Fiberlink Discs, 877 CPU -------- PERFORMANCE PLANS: We will be well positioned to have a strong LADDIS (SPEC replacement for Nhfsstone) in the Fall when it becomes available. HP will attempt to publish LADDIS performance information for both our series 800 and 700 HP/UX systems as soon as possible after acceptance of LADDIS. COMPETITIVE PERFORMANCE: FYI: Here is a short collection of competitive information from GSY. If you receive competitive information on NFS from your customers/peers, please notify the Network Sales Center at 447-4444, so that they can provide the information to others who may need it. AUSPEX: Using Nhfsstone--standard mix, 8K writes. Similar to 1.22, Model BM # LANS # disk IOPSs Delay NS 5000 V1.22? 8 LAN 16 SCSIdisk 1100 73 (est from graph) NS 5000 V1.22? 8 LAN 16 SCSIdisk 980 44 " NS [35]000 V1.22? 4 LAN 8 SCSIdisk 560 42 " Note: Auspex has begin showing pre-LADDIS results of > 1300 IOPS for the NS5000. They are also quoting pre-LADDIS results for their new system, the NS5500 of over 2000 IOPS. HP will begin quoting LADDIS numbers as soon as the bench- mark becomes officially accepted by Spec. SUN Using the known to be "lighter" 2.03 version SS6x0MP v2.03 4 copLAN 782 37 SS490 v2.03 4 copLAN 725 42 COMPETITIVE TIP FOR SUN: Note: Sun has also announced 1000+ estimated results for the just announced model 41 system. These results are on net yet available hardware. HP's 1000+ IOPS results were measured on released and shipping hardware and software using available NHFSstone 1.22. ADDITIONAL PERFORMANCE INFORMATION: Full performance briefs are available for the series 700 and 800 systems. For additional information on NFS performance, please contact the Network Sales Center, at telnet 447-4444. NFSPERF May 26, 1992