Myths and legends When database folks gather to sup ale and chat, as they invariably will at Oracle's OpenWorld in San Francisco this week, talk occasionally turns to benchmarking. As soon as it does, someone will say: "Of course the trouble is you can't trust the vendors. Wasn't Oracle caught cheating at the TPC benchmarks?"
The public has a right to know the truth...
The Transaction Processing Performance Council is a non-profit corporation that defines its role in life with commendable clarity. The organization was founded in 1988 to "define transaction processing and database benchmarks and to disseminate objective, verifiable TPC performance data to the industry".
By devising tests that ensure all systems are performance tested in the same way for the same criteria, the TPC - strangely the second P seems invariably silent - aims to provide the fairest possible comparison between systems. The TPC's earliest benchmark was TPC-A for On-Line Transaction Processing (OLTP) systems: the two most often cited today are TCP-C (a heavily revised and updated version of TCP-A) and TCP-H for decision support systems.
Each benchmarked system is given a score; the TPC-C quotes transactions per minute (tpmC) and also a Price/tpmC. These figures give customers and industry commentators a way to make meaningful comparisons between systems - and manufacturers take an even keener interest in the results.
In April 1993 the Standish Group claimed Oracle had incorporated an extra "discrete transaction" option into its product that would let it perform particularly well against the TCP-A benchmark but offered no noticeable benefit to customers. Standish opined that, by adding this option, Oracle was putting forward for testing a "benchmark special" and that in doing so the company had violated the spirit of the TPC.
Oracle vigorously refuted these claims, saying the system tested adhered fully to the published benchmark specifications. The company also pointed out that since these specs made no mention of benchmark specials, it could not possibly be in violation of anything. And Oracle was right: the TPC had no rules to exclude special editions of software for testing.
It was an area the TPC suddenly felt moved to address and by September that year a clause had been added to the specifications that prohibited benchmark specials. It came into effect in June 1994. Oracle did not test its discrete transaction-enabled version against the beefed-up benchmarks and by October had withdrawn from its literature any mention of the benchmark score gained in tests on that version.
So was Oracle's behavior legal, decent and honest?
It was certainly legal: the TPC was aware of the Standish Group's claim but never formally discussed whether Oracle had put forward a benchmark special or made any decision about it. Since the TPC never discussed the issue of "cheating" I think it is unreasonable to accuse Oracle of cheating.
Deciding whether decency and honesty were best served, though, is a matter of opinion and best left as an exercise for the reader.®