James Sha, vice-president in charge of Belmont, California-based Oracle Corp’s Unix division, has raised objections to Pyramid Technology Corp’s performance claims for the Oracle Banking Benchmark running on its recently-announced top-end T-series database servers, (CI No 1,476). Sha says that Pyramid derived its figures from running an unscaled benchmark methodology that results in a very small account table – 10Mb – that can be completely contained in memory. This size, he says, does not exercise the disk subsystem and is unrepresentative of actual user applications. A scaled benchmark would require an account table of at least 2Gb, ensuring that data must be retrieved from disk. He says that other performance comparisons Pyramid made with Sequent Computer Systems, IBM and DEC systems – were based upon results from the announcement of Oracle 6.0, over two years ago, and are extremely misleading. Furthermore Sha says Pyramid’s cost per TPS calculation did not conform to the industry practice, which should include enough disk to contain a sealed database – 2Gb or more – 90 days of transaction history 25Gb or more, and hardware and software maintenance costs for five years. Pyramid, he claims, based its costs only on the hardware cost of the system that ran the Oracle benchmark. Ian Couper, Pyramid UK’s marketing manager has responded by saying that although the performance results used at the launch were two years old, in all cases – except for the Sequent figures, which are from this year – they are the last results that are publicly available. He says an independent benchmark of the T series is now being carried out by Codd & Date, which will also include other relational database suppliers. While Oracle looks to have been pushed into raising objections by irate competing manufacturers that also use its database software, both Pyramid and Oracle maintain that they are still the best of friends behind the scenes – and anyway, the Transaction Processing Performance Council is intended to put an end to all these unproductive arguments.
