Black, Alain wrote:
It sounds more like why when benchmarks are run, they are run 3 or more times with the first run being thrown out.
This is due to the initial loading of data (reading files off of disk into cache) that must be done. The next run will not have as much disk I/O and if the test is small enough, could actually be reading straight from memory, which is significantly faster than disk.
After a reboot, or being powered off, the cache is clear, so all data needs to be read into memory again.
-Alain.
Alain, I totally agree with you, but Sheo Shanker Prasad stated that under 9.1 and the old mobo, he got consistently the same execution time... So, now with 9.3 and the new mobo, there is a different behaviour... How do you explain that ? cheers, fabrice