By John Taschek, PC Week
I was staying at a Motel 6 (don't ask) near Palm Springs, Calif., for the Intel Developer Forum when I got the heads-up that Microsoft would "drop a bomb" on Sun Microsystems.
That bomb turned out to be SQL Server 2000's performance numbers, which not only rocked Sun, but IBM, Oracle and Sybase. They even gave me a stir, and I don't believe anything anymore.
Buried beneath the Windows 2000 launch in San Francisco was the news that Microsoft somehow tweaked a beta of SQL Server 2000 enough to achieve record-breaking Transaction Processing Performance Council TPC-C numbers (see www.tpc.org). Ordinarily, this wouldn't be a big deal. This time, however, SQL Server generated over 200,000 tpm (transactions per minute) for just about US$19 per transaction.
This blows away every number in the chart. Of course, TPC-C is reaching the end of its usefulness, but SQL Server 2000 hit record numbers on other benchmarks as well, indicating that a) Microsoft engineers really know how to tune an engine, b) to get these scores, Microsoft cut out things that other people may think are necessary (think of driving a Chevy Vega with no roof and no doors), or c) the product really is fast.
Clearly, it's all three. Microsoft seems to be able to double SQL Server's performance every 18 months, and this can't all be attributed to improvements in hardware.
Then there's the issue of Microsoft tweaking the platform so much that it no longer resembles a real system. Microsoft admitted that this was a nonpublic beta of SQL Server 2000. It's also obvious that Microsoft's choice of a transaction processing monitor (COM+) isn't in the same league in reliability and extensibility as BEA Systems' Tuxedo, for example. Vendors play around with different configurations all the time—that's what they do. But take it from me—these numbers are as credible as we're going to get. That means that SQL Server 2000 is amazingly fast. It sure blew the doors off Oracle8i. Oracle tried to upstage Microsoft with a Superman-themed press release, boasting about its 100,000-tpm number. Ouch. Bad timing.
The competitive database vendors may scoff at the concept that SQL Server 2000 might be the fastest relational database. They might say, We could put rocket fuel in that Vega and it would fly ... once. I'll bet, however, that when marketing officials are out of the public eye, they are very afraid.
So what does this all mean? Back when databases were important in their own right, numbers were critical. Now it doesn't matter. Those who manage databases will absolutely, positively never change. Meanwhile, new companies almost always choose Oracle. It's a fact of life that most dot-coms are run by investors who like to take the road most traveled, which mostly means Oracle.
So the only thing for Microsoft to do is to give away free stuff. When SQL Server 2000 ships, it will include an integrated data mining engine. This could be huge, since data mining engines cost upward of several hundred thousand dollars. Microsoft is trying to do with data mining what it did with OLAP: make it free. There will be more, Microsoft said, but where will it all lead?
Is Microsoft helping to revive the database industry?