> For what it is worth I did some trials in matching page size to the > underlying virtual memory page size and was surprised to find that I did > not measure any significant performance change. My guess is that it is > the structure of the data in your application which would be sensitive > to page size. > JS
I will concur with this :) Our tests were done by three different people (Who all wrote their own test programs and had their own ideas about what would be the best way to test this) We discovered it made no difference at all in one case, and a small, yet noticeable difference in the best case. Someone already replied here saying a 15% speed increase is probably the best case. I agree that's probably the best result you'll get, but add a rider that you are unlikely to get that much. You will probably get better savings fine tuning your code and algorithms
