On April 24, the Itanium-powered HP Superdome moved to the top of the TPC charts for non-clustered servers, with a TPC result of 658,277 transactions per minute (tpmC) at a cost of $9.80 per transaction. Then, last week, IBM's eServer pSeries 690 Turbo achieved 680,613tpmC at a cost of $11.13 per transaction.
The TPC website revealed that while HP's system is indeed cheaper, it based its configuration on HP's SCSI-based MSA storage subsystem rather than enterprise-class EVA products. IBM pSeries product manager Simon Roberts said, "I do not know of any user who would run a SCSI-based large-scale San in their datacentre," suggesting that HP's configuration was unrealistic compared to IBM's.
The breakdown of the performance figures shows that, in terms of software licensing, the DB/2 and AIX configuration for the 32-processor IBM system, is more than $500,000 (£310,000) cheaper than 64-bit Microsoft SQL Server and Windows 2003, which run on the HP Superdome.
Defending the cost of the software on HP's system, Richard George, business-critical server business manager at HP, said, "DB/2 is heavily discounted due to the ongoing price war between IBM and Oracle. What IBM gives away on DB/2 it makes back on its storage."
Neither of the benchmarked systems is a realistic option for users, said analyst Rakesh Kumar of Meta Group. "I wish suppliers would realise that users are looking for integration and support in their hardware, and not simply TPC performance," he said.
Kumar warned users to take the figures with a pinch of salt. "The hardware suppliers do play tricks," he said.
Ovum analyst Gary Barnett said, "Buying on the basis of TPC benchmarks is as meaningful as buying a road car based on which manufacturer wins the Grad Prix.
"You would be mad to base your purchasing decision on the results of any benchmark, but they can serve as a guide to help you narrow down choices and identify likely candidates."