This article is more than 1 year old
Nemesis looms over TPC-C benchmark
Expensive and oudated
TPC-C - that venerable server-come-database benchmark from the Transaction Performance Processing Council - looks set to be pensioned off at last with the recent announcement of a new OLTP (OnLine Transaction Processing) benchmark specification, TPC-E.
The new benchmark is aimed at moving the touchstone of database and server performance into the world of today rather than yesterday. This is being achieved by shifting the fundamental premise on which the specification is built to a model more fitting to current business needs.
TPC-C was built around a model of a typical database system needed to run a parts warehouse. With TPC-E the model has been shifted to a model of a typical brokerage operation. This should give a better simulation of the modern real-world transactions that take place over the Internet – including the time delays and fractures that can occur between distributed contributors to the smooth running of a transaction. In operation, the benchmark goes through the process of interacting with the financial markets to execute customer orders and update the relevant account information files.
The Council has also aimed at getting more reality into the benchmark by creating `real’ names, addresses and business details for the test database from data culled from US census and New York Stock Exchange information.
TPC-E is also scalable, so that the number of customers can be varied. This should expand the benchmark’s usefulness in allowing database and server vendors to pitch at small and mid-range market sectors specifically, and at a price that smaller vendors can afford. The price of conducting benchmark tests has been one of the recent criticisms of TPC-C, so much so that only the major vendors now attempt it.
One reason for this is the suggestion that results can be affected by factors such as the number of disk drives used, rather than the total disk capacity. So, more small capacity disks can mean a better test result. But it also means an impractical installation configured specifically for the test, sometimes with several thousand disk drives and a price tag running well into the millions. This is decreasingly relevant to the average business requirement, though the benchmark results are still seen as good guidance.
TPC-E is still currently in draft as Version 1.0.0 and more information, together with the detailed specification, can be found here. ®