Perception vs. Reality

Coming on the heels of Oracle's latest $1 million challenge, the recent Transaction Processing Performance Council (TPC) Web site reorganization seems to be further evidence that Oracle is more worried about Microsoft and SQL Server 2000 than you might expect. I discussed the fine print of Oracle's million-dollar discount in "Paying to Play," February 2001. However, the new TPC site reorganization is an altogether different development.

The recent slew of high-end TPC-C scores achieved by Microsoft and IBM have threatened to push Oracle out of the top 10 performance spots. In February 2000, for the first time ever, SQL Server topped the TPC-C performance list with an early clustered implementation of what we now know as SQL Server 2000. After several months of nit-picking, an unnamed TPC member managed to get the Microsoft scores withdrawn because the cluster (like all the other previously recorded clustered systems) didn't allow for primary key updates. However, this short-lived advantage backfired on Oracle after IBM used a clustered DB2 implementation to easily claim the top TPC-C mark. To no one's surprise, Microsoft quickly followed IBM's mark by submitting five new SQL Server marks in July 2000, which resulted in SQL Server 2000 once again sitting at the top of the TPC-C score card and claiming four other top 10 spots. In October, Oracle 8i managed to climb into the number 5 spot with a mark of 220,907 transactions per minute (tpmC), less than half of the top SQL Server score of 505,302 tpmC.

However, in Oracle's case, I guess if you can't beat 'em, you can at least make your score look better. The TPC Web site now sorts the TPC-C scores by clustered systems and nonclustered systems, allowing Oracle to be on top of at least one TPC-C ranking. This change might make Oracle look better, but I'm pretty sure that the SQL Server team members are the ones who feel better. Given the recent availability of Windows 2000 Datacenter Server and the SQL Server 2000 enhancements that let SQL Server take advantage of multiple processors, you know it won't be long before you see SQL Server climbing through the ranks of the nonclustered systems. Of course, I don't think that TPC-C scores directly correlate to the performance that you might see from a particular database implementation in a real-world situation. But the scores give a reliable prediction of the high-end scalability that a particular platform can provide.

You can check out the recent TPC changes at http://www.tpc.org. While you're checking, note that SQL Server 2000 and SQL Server 7.0 claim all 10 of the top price-to-performance slots. Anyone familiar with SQL Server knows the great performance that SQL Server delivers on relatively modest hardware, and these results are further proof. That feels pretty good.