Let Tr be the minimum response time for an operation of any sort, Tf be the time a handle must sit before being moved back to the free list, and Ntot be the total number of handles available on a server.
The pathological case would be one where a caller
This results in the largest number of handles being unavailable due to sitting on the overflow_list. Call Npurg the number of handles waiting in ``purgatory'' ( waiting for Tf to pass)
Npurg = Tf/Tr | (1) |
Fpurg = Npurg/Ntot | (2) |
Fpurg = Tf/(Tr*Ntot) | (3) |
We should try to collect statistics and see what Tr and Npurg end up being for real and pathological workloads.