emv@ox.com (Ed Vielmetti) (03/05/91)
I'm looking for some guidance on measuring the performance of a system which seems to have a lot of contention for a piece of (system V style) shared memory. First help would be any sort of shared memory benchmarks that anyone has so that I could get reasonable quantifiable numbers on shared memory performance (i.e. something to convince me that operations on a Decstation 5000 are nn% faster than on a Dec 3100). A look through Eugene Miya's benchmarking list didn't yield anything obvious, but if some piece of a bigger benchmark suite has some tools that would help quite a bit. Are there tools better than ipcs(1) for examining the usage of shared memory? Nothing obvious shows up along the lines of vmstat, netstat, iostat et al. Thanks as usual for any help. -- Edward Vielmetti emv@ox.com
mah@dec1.wu-wien.ac.at (Michael Haberler) (03/12/91)
In article <EMV.91Mar4113610@poe.aa.ox.com>, emv@ox.com (Ed Vielmetti) writes: |> |> I'm looking for some guidance on measuring the performance of a system |> which seems to have a lot of contention for a piece of (system V |> style) shared memory. I'd watch for behaviour of programs with several shared memory segmments attached. On some architectures, there's an upper limit on how many segments are actually attached, and if more than that number of segments are used, they are detached and reattached by the kernel on the fly - with page faults and a kernel trap. That can mean a big difference in performance. I could imagine that that limit might be decreased by the use of shared libraries. - michael