New Database Benchmark Closes Loopholes That Made Tests Unrealistic

Mar 23, 2007 (08:03 PM EDT)

Read the Original Article at

Build a benchmark to compare product performance, and someone's going to find a loophole. But recent work to update the 15-year-old standard for comparing performance of databases should make the results a much more meaningful reflection of today's complex transactions.

The benchmark--TPC-E, from the vendor-sponsored Transaction Processing Performance Council--uses real publicly available data, such as names from the U.S. Census and trading data from the New York Stock Exchange, to arrive at a transactions-per-minute measure. It requires handling several different types of data in a multistep transaction.

TPC-E sought to close loopholes in its predecessor, TPC-C, which specified steps of a transaction to be executed by a system but left implementation to each vendor. That led to shortcuts, such as tests that searched for masses of user names that all began with the same two letters, or tests that retrieved highly similar data from just one section of the database. Those shortcuts yielded results better than any real user could expect.

The old benchmark mimicked a straightforward retail order from a warehouse. TPC-E replicates a stock trade through a brokerage house, with several complex steps along the way.