We recently had an issue about the usleep call taking a lot longer than 1 us for one user. Probably related to the hardware if the usleep implementation is hardware dependent.
Anyways I'd like to point out that we commented out the usleep function, actually removing every implementation of this function, and it did not influence our results. No problems at all. Only 16x speed-up, 5KB / sec right now.
Actually the function is called between IO memory calls to ensure IO completion. This is not required, since the system blocks IO calls until completion.
Regards, Raul
since the system blocks IO calls until completion.
system means operating system here.
The calls to the wait function have been removed in SVN.