Unisys gets a first
The only TPC-E entrant, so far, `wins’
I feel strangely sorry for Unisys, even though it has just won a race that is, arguably, well worth crowing about. But you see, it is the first company to run the new TPC-E benchmark that replaces the aged and venerable – and to be honest somewhat discredited – TPC-C benchmark.
It is, together with its partner in the process, Microsoft, the first to have generated any real numbers for the benchmark – and that is the problem. It is a bit like me running (running? Oh that is a bad-taste joke. Crawling or staggering would be far more appropriate words) the 125 metre `dash’ in 2 minutes 45.72 seconds. If I am the only one in this competition then maybe I did really well – who can tell?
So the Unisys results - 660.85 tpsE (transactions per second for the TPC-E benchmark) and a price/performance rating of 1,777.18 $/tpsE – may be stunning, merely good, or rather laughable. Only time will tell. In practice, it won’t be until there are a good half dozen results out that users will be able to get a feel for what the numbers actually mean.
This is not really helped by the fact that the announcement from Unisys makes no mention of a factor that could be quite important in assessing future TPC-E results. TPC-E is designed to mimic a financial brokerage operation, and can be scaled in terms of the number of named customers that are handled. So is this a Unisys `brokerage’ with one customer or one thousand? And how might that impact the numbers anyway?
One thing that is certain is that the rig chosen for the test, an ES7000/one Enterprise Server system running Microsoft Windows and SQL Server, now becomes the primary target for both server vendors and database vendors. So expect to see a small glut of competitor results appear before the Unisys system goes on sale towards the end of September.
I think it is also possible that developers in the database arena may find themselves contemplating ways in which the databases and infrastructure can be tuned to get the best possible results – if only to allow them to `reverse engineer’ the results into something meaningful for their own installations. After all, no matter how valuable the fundamental numbers generated might be with a bit of understanding, once the marketing department had done with them it will come down to smoke and mirrors as usual.
But while the winner’s laurel goes to Unisys, rest assured that the company now risks being trampled in the inevitable rush to beat its figures.