I'd think that the falloff would be more sharp in that case. Since Linux by default has such low affinity, if you have too many processes you're probably losing out on coherency of memory and hard drive accesses.
I'd think that the falloff would be more sharp in that case. Since Linux by default has such low affinity, if you have too many processes you're probably losing out on coherency of memory and hard drive accesses.
Yes, that makes sense. In that case, there would be little (gradual) difference in performance between, say, 6 or 16 separate threads, as long as it has to start buffering and swapping them. The processes themselves probably share a single codespace, so that would make little difference.