Baseball is great for statistics. Every game is like a miniature numbers factory, generating batting averages, on-base percentages, earned-run averages—you name it. For fans, these numbers constitute the game outside the game: when the lights go out and the players have all gone home, the fans continue the battle using statistics for ammunition.

 

In the server world, performance benchmarks can be like that too. People love to debate them. Like baseball, the numbers tell a story—looking at a series of results you can see come-from-behind victories, heartbreaking losses, and the rise and fall of competitors. So as a benchmark junkie, I was keenly interested in yesterday’s bumper crop of performance results published by HP, all using the SAP SD 2-Tier benchmark as their yardstick.

 

All of the results are based on the forthcoming Window Server 2003 Enterprise x64 Edition—so they are among the earliest proof points around this product in the public domain. Secondly, they all use the x64 version of SAP R/3 4.7E, which is totally cool in my book because it gives a sneak peek at some of our application support. Third, they all use the 32-bit version of Microsoft SQL Server 2000. Yes, that’s right—the 32-bit version. Why does this please me? Because the whole promise of this platform is its ability to run existing 32-bit applications as well as new 64-bit stuff, all with great performance. So what could be better than an audited set of performance tests that combine 32-bit and 64-bit software?

 

Hopefully by now I’ve piqued your interest as well, so I’ll cut to the chase and tell you how this all panned out. The best score was 1017 concurrent users on an HP DL585 with four Opteron 852 (2.6GHz) processors and 16GB of RAM (a round of applause, please). That’s quite good. In fact, it’s the best result among all 4-way systems except for one using Power5 chips (and unfortunately the SAP benchmark doesn’t include price-performance, or there would be an interesting comparison there, I’m sure). Using the same software stack, HP also claimed top honors among 2-ways using an HP DL370 G4 with two Xeon DP (3.6GHz) processors and 16GB of RAM, which supported 597 concurrent users. Putting the two results together shows we’re getting good scalability in addition to raw performance. You can see all the results for yourself at http://www50.sap.com/benchmark/.

 

One of the side stories to all of this is the fact that HP now has Opteron-based systems in their ProLiant 300 series, 500 series, and BL series, so the range of customer choices continues to expand. Fujitsu-Siemens has also published results using the x64 version of Windows, and Sun has published a Windows result using their V20z Opteron model.

  

So what to make of all this? I like to step back from the speeds & feeds and look at the broader picture. First, we know that there is merit to at least some of the hype and excitement around this new platform—in other words, it does what it’s supposed to do: improve performance and scalability and support both 32- and 64-bit apps. Second, we know that leading vendors are getting behind it early, and that bodes well for customers—the more vendors offering x64 solutions, the better the choices, the broader the knowledge and expertise, etc. Third, we know that Pedro Martinez walked 37 batters after pitching 213.1 innings in 1999, which means…oh wait…never mind, I was thinking of something else.

John Borozan