The Big Win2K Surprise

• 03/08/2000

At the Feb. 17 launch keynote in San Francisco, Microsoft Corp.'s chairman and chief software architect Bill Gates started yammering on about Transaction Processing Performance Council benchmark results.

Yawn. ENT had covered the recent eight-processor Compaq ProLiant/Windows NT results using an Oracle Corp. database in a six-node Oracle Parallel Server configuration. There was no surprise when Gates displays those numbers on the big screen to show Microsoft in the raw performance top 10 -- for the first time breathing the rarified air that was the exclusive domain of Unix/RISC vendors.

Suddenly, Gates showed another set of TPC numbers that caused my jaw to drop. Microsoft had just gotten audited results that put its systems at the top of the raw performance list. Above IBM Corp. Above Sun Microsystems Inc. Above any system running an Oracle database.

Seems the crew in Redmond can keep a secret after all. Of course it didn’t hurt the secret-keeping process that Microsoft got one of the record-shattering results audited the day before the Windows 2000 launch, and the other one at 4:30 the morning of the launch.

One system was eight nodes of Windows 2000 Advanced Server running beta versions of SQL Server 2000 on eight-processor Compaq ProLiant servers. The other system was identical, except it ran across a dozen nodes and brought a 50 percent performance improvement over the eight-node system. Both systems relied upon a previously unannounced feature of SQL Server 2000 that allows for performance clustering (For more detail on the systems, see the story at http://www.entmag.com/displayarticle.asp?ID=380052246PM). The systems are listed as being available Aug. 1.

Do I need to mention that both systems cost considerably less than the Unix/RISC systems they outperformed? We are talking Wintel here.

There’s an old saying about lies, damned lies, and statistics. Analyst Dan Kusnetzky at International Data Corp. says the phrase "benchmark results" deserves a place in that saying.

These benchmarks don't consider cost of management. The IBM Corp. benchmark Microsoft crushed was achieved on a single machine with 24 processors. That’s considerably less difficult to manage and maintain than a bleeding edge cluster of eight or 12 SMP machines cobbled together on beta code and the spit and sweat of Compaq’s and Microsoft’s brightest.

Depending on your perspective, benchmarks involve cheating at worst, or "tuning" at best. Still, the fact remains that Microsoft previously could never cheat, er, tune, anywhere near as well as IBM and Sun. Now Microsoft can keep up with the best of them.

Competitors like Oracle’s Larry Ellison and Sun’s Scott McNealy used to be able to say that when it came to scalability, Microsoft couldn’t get there. Now, they can only say they don’t like the IIwayII Microsoft gets there. But that’s a very different kind of argument. Can you say religious war?

This benchmark will be broken shortly. Microsoft admitted as much on the day they announced it. There’s a good chance it’s been broken by the time you read this column. IBM and Sun both scale up to higher levels of performance on multiprocessor machines than Microsoft is currently capable of doing. Microsoft solved the problem by scaling out with many machines, but there’s nothing to stop competitors from doing the same.

Nonetheless, Microsoft is firmly established as a contender. The company’s ability to scale with Windows 2000 is absolutely as good as anyone could have reasonably hoped.

Fortunately, things are just getting started. Unisys? Can we see some 32-way results on Windows 2000 Datacenter Server please?

=

[Infobox]A New Seat on the Bench

Source: Transaction Processing Performance Council (www.tpc.org)