Performance measurement is not a simple task. There is more than one trap to walk into. An important one is that in the first execution data may be on disk, and the next it may be in cache. You need to decide what you want to measure. If you want to load with data on disk, you need to do DBCC FREEPROCCACHE before each run - but don't do this on a server where other people are working.
And, well, if you test on a server where other people are working, their work will affect your measurements. If you measure on your own machine, beware for other things that happens on the server.
Erland Sommarskog, SQL Server MVP, esquel@sommarskog.se
No comments:
Post a Comment