Benchmark retest favours HP server

Hewlett-Packard and Sun Microsystems have recently upped the ante in the benchmarking stakes with impressive results for the...

Hewlett-Packard and Sun Microsystems have recently upped the ante in the benchmarking stakes with impressive results for the retested Superdome server and the latest Ultrasparc chip respectively, writes Eric Doyle.

Superdome, HP's thoroughbred 9000 series server, has been retested to the TPC-C (Transaction Processing Council) benchmarking standard to reveal a transaction processing figure double its previous result.

The first results showed little difference between Superdome and comparable systems from IBM and depressed the potential sales of the system that had been predicted pre-launch.

The company is now trying to prove, in court, that the original results were deliberately sabotaged by its former employee Hock-Beng Lim. The charges against him imply that he was paid by an unnamed third party to ensure a less-than-impressive result.

The latest results are nearer to the degree of performance that HP expected, but failed to achieve, when the system was launched at the beginning of this year.

The TPC-C tests are well respected in the industry as an index to the performance of a hardware/database combination and are devised to emulate a real-world online transaction processing environment.

The goal is to reveal comparable results regardless of the processor and operating system in use. The resultant report reveals an index figure in transactions per minute (tpmC) from which a price/ performance indicator, in terms of dollars per transaction, can be derived.

Critics of the TPC methodology focus on the fact that the test does not truly represent hardware performance because there is a free choice of the database used.

Whatever the view of the comparability of individual system results, HP's latest figures are, without doubt, impressive. The system differs from previous tests because it uses the recently-released PA-8700 Superdome with the Oracle 9i database in a configuration that will not be available until May next year.

In the last test on the $400,000 (£285,000) system, taken in May this year, the benchmark was 197,024 tpmC, which represents $43.25/tpmC but this has been increased to 389,434tpmC - $21.24/tpmC.

This is a vastly superior result compared to IBM's best result so far, taken in April using an IBM p680 Unix server with the older Oracle 8 database, where the figures were 220,807tpmC and $34.18/ tpmC.

However, HP is still trailing behind Fujitsu's Sun Solaris- powered Primepower 2000's August result using a Symfoware Server Enterprise Edition database, which together gave a tpmC rating of 455,818 but a less favourable £28.58/tpmC result.

Sun, which currently has only one test result under TPC - Sun Enterprise 4500 at 67,102tpmC, $25.85/tpmC - prefers to promote its Spec (Standard Performance Evaluation Corporation) results for compute-intensive CPU performance.

Using the CPU2000 test suite, Sun's latest Sparc processor performed surprisingly better than previous iterations.

The Spec tests result in scores, known as a mark, being given under various categories of test. This allows a comparison to be made between processors according to which scores higher and the main categories are compute power when handling integer (whole) numbers and when using floating point (decimal) numbers.

The Ultrasparc III Cu 1050 processor revealed a results of 610 for Spec integer and a Spec floating point of 827. This shows a notable improvement over earlier iterations of the chip of 32% and 72% respectively.

Read more on Business applications