Windows-Itanium Still Lags Big Unix on SAP Tests

by Timothy Prickett Morgan

Operating system and platform maker Microsoft and Itanium server partner NEC have rolled out a new benchmark test result on the SAP ERP suite's Sales & Distribution (SD) test, and the two say that the combination of 64-bit Windows and 32-way Itanium 2 processing has broken the records for 32-way processing. While technically true for Windows platforms, there is a lot of big Unix iron out there that can meet or beat the performance that Microsoft and NEC are showing off with this two-tier SAP SD test result.

And it stands to reason that even a 64-bit Linux environment running on the same NEC "Asama" Express5800/1320Xe server would yield very similar results on the SD test, too.

There are two types of SAP SD tests. The two-tier variant puts the SAP database and the application servers that bang out transactions against it on a single machine, while the three-tier variant uses external application servers and only measures the performance of a single, central database server.

On the two-tier test, a 32-way NEC Itanium server configured with 128 GB of main memory, 931 GB of disk capacity, and 32 of Intel's 1.6 GHz/9 MB cache Itanium 2 processors was able to support 5,210 SD users. That setup processed 1.572 million dialog steps per hour (dialogs are portions of SAP transactions) at an average response time of 1.93 seconds. This server was running Microsoft's Windows Server 2003 Datacenter Edition and SQL Server 2000 database, and it ran at 94 percent of peak processor performance. That is a little less than twice as much performance as the 16-way Windows-Itanium servers that have been tested by Hewlett-Packard, Groupe Bull, and Unisys using 1.5 GHz Itanium 2 chips, and a little more than twice as much as the 16-way xSeries 445 servers using 32-bit Xeon MP processors from IBM running the same Windows stack.

NEC's performance is not too shabby, of course. But big Unix iron is not exactly rusting because it moves too slowly in the server market. For one thing, HP was able to show considerably more performance on the SD test with Integrity servers outfitted with 1.5 GHz processors running HP-UX, pushing performance to 2,880 SD users running HP-UX 11i and the Oracle 9i database, compared to 2,160 users running Windows and SQL Server.

More significantly, a 32-way Fujitsu-Siemens PrimePower 1500 machine tested last fall on the same two-tier SD test using 32 of Fujitsu's 1.89 GHz Sparc64 V processors with 3MB of on-chip cache was able to support 5,200 SD users running Solaris 9 and using Oracle 9i. So the Windows-Itanium box just barely squeaked by the Fujitsu-Siemens Unix server by 10 users. This is not exactly the record-breaking performance that Microsoft and NEC are trying to portray. And it is important to note that the NEC Asama server is out of gas at 32 processors, but that the PrimePower machine can scale to 128-way processing and handle maybe somewhere around 16,000 to 18,000 SD users with the same software as was run on the 32-way PrimePower Unix box. NEC has shown the top two-tier Windows performance, but HP has yet to demonstrate how its 64-way or 128-way Integrity boxes using Windows do on the two-tier test. (It has run SD benchmarks on the 64-way machine on the three-tier test, however.) It is hard to imagine a 64-way Integrity box running Windows and SQL server doing much better than 10,000 SD users on the two-tier test, and even with the mx2 dual Itanium modules that HP created when Intel could not ship its own dual-cores processors, HP could probably only boost Windows-Itanium performance by a little bit more since the mx2 modules have their clock speeds geared down so you can put two of them in a single Itanium 2 slot. Real dual-core "Montecito" Itanium processors are not due on the market until early 2006.

All of this means that it will be a long, long time before the Windows-Itanium combination comes close to the raw top-end performance of IBM's p5 595 Unix servers. Last fall, IBM was showing off a 64-way p5 595 running its AIX V5.3 Unix and its DB2 8.2 database, which was able to process over 6 million dialog steps per hour with an average response time of 1.92 seconds. That worked out to 20,000 SD users in the two-tier test on a 64-way machine running at 97 percent of peak capacity. That p5 server was equipped with dual-core Power5 processors running at 1.9 GHz; every two cores shares a 1.92 MB L2 cache and a 36 MB L3 cache. If Microsoft wants to get similar performance for its Windows platform, it basically has one choice for the foreseeable future: port Windows 2003 or Longhorn Server to the future Power5+ or Power6 platforms from IBM. Intel is not going to deliver multicore Itaniums with the kind of performance IBM can bring to bear. It is that simple.

An honorable mention also goes out to Sun Microsystems, which last summer supported 10,175 users on its top-end Enterprise 25000 server running Solaris 9 and Oracle 9i. That machine was configured with 144 UltraSparc-IV processor cores running at 1.2 GHz, however. That high core count is why, in part, Sun had to partner with Fujitsu-Siemens to deliver more powerful SMP servers with fewer processors. With a lot of software still based on processor counts or core counts, the E25K is only preferable to a PrimePower box if customers have enterprise licenses to systems and applications software and they need threads more than they need raw single processor performance.