Unisys Closes in On Unix/RISC Performance Mark
- By Scott Bekker
slid in for the silver on SAP's proprietary benchmark for scalability on Tuesday, narrowing the gap between "big iron" Unix/RISC systems and "big iron" Wintel systems.
The result on the 32-processor ES7000 server running Windows 2000 Datacenter Server and SQL Server 2000 Enterprise Edition, takes Unisys from fourth place to second on SAP's Sales & Distribution (SD) benchmark.
Unisys achieved 20,000 concurrent users on the ERP vendor's benchmark by replacing the 700 MHz Pentium III Xeons it used in its last run of the benchmark with Intel's new 900 MHz PIII Xeons. The figure surpasses the 19,360 users posted by Sun Microsystems Inc. with a 64-processor, 64-bit system running Solaris and Oracle Corp.'s 8.1.5 database.
However, Unisys' assertions that its "ES7000 surpassed the best performance of any Sun/Oracle system" is somewhat misleading.
Unisys trails a Fujitsu-Siemens system that ran the benchmark using Sun's Solaris operating system and Oracle's database a healthy 13 percent faster than Unisys' latest effort at 23,000 concurrent users. Fujitsu-Siemens sells Primergy brand systems that compete with Sun's Starfire hardware but run the Solaris operating system.
Fujitsu-Siemen's result is also much more recent. That company posted its benchmark in November. The Sun benchmark Unisys cites is about 15 months old.
The benchmark run is Unisys' fourth effort on the proprietary SAP benchmark. Unisys chose the benchmark to demonstrate the scalability of its relatively new "Intel mainframe" using 16 processors, then 24 processors and more recently 32 processors.
Although Unisys' marketing of the benchmark result continues to claim a price advantage over Unix systems, the company's choice of benchmark hides the exact cost advantage from customers. A weakness of the SAP benchmarking process is that it fails to show customers specific pricing information.
Industry observers are waiting for Unisys to publish on the industry-standard Transaction Processing Performance Council's TPC-C benchmark to find out how Unisys' server really stacks up on price/performance against Unix/RISC systems.
The 32-processor database server Unisys used is merely one part of the massive configuration Unisys used to generate the benchmark. Feeding the database server were 92 application servers, each with eight 700 MHz Xeon processors, and 623 GB of storage.
By comparison, Fujitsu-Siemens had 161 application servers feeding its database server, although all but one of those servers had four processors. For more detail about the benchmarks, see SAP's Web site.
TO THE EDITOR:
I was impressed by the level of effort devoted by Scott Bekker to researching the facts behind our latest SAP performance benchmark ("Unisys Closes in On Unix/RISC Performance Mark," June 12).
I think it is worthwhile however to respond to the assertion that our claims to have surpassed the performance of any best Sun/Oracle are "somewhat misleading." I might agree if the Fujitsu-Siemens server that achieved higher performance than that recorded by Sun was in fact identical to, and a rebadged version of, the Sun E10000.
However, the Fujitsu-Siemens PRIMEPOWER Model 2000 is in fact a higher-performance extension of the E10000 server architecture. There are substantial differences between the two servers, including the fact that Fujitsu-Siemens develops and manufactures its own SPARC processors.
The article is also correct in observing that the SAP benchmark methodology does not provide for price/performance analysis and comparison. However, Aberdeen Group has done its own analysis based on public price information and concluded that the price per user of the Sun and Unisys systems as tested by the benchmark were $121.51 for the Sun system and $44.16 for the Unisys ES7000. The Aberdeen report is available on our website.
Scott Bekker is editor in chief of Redmond Channel Partner magazine.