Michael Clark wrote:
* A stackable performance statistics vfs module implementation that does timings on filesystem io calls.
BTW I did some timings on linux for the wrapped apr_stat unix socket sendmsg/recvmsg ping/pong calls to the privsep process and a OS cached stat was about 22 microseconds vs 4-5 microseconds for a regular direct apr_stat call.
At this level (45,000 calls per second), we would most probably be seek IO bound in real life anyway (in our use case the mod_ssl encryption load would by far dominate).
