Bill Sommerfeld writes: > On Fri, 2007-09-14 at 13:06 -0700, Alexander Kolbasov wrote: > > Just being Devil's advocate. I don't doubt that you should be able to > > improve the measurement significantly. I just think that the notion of > > CPU utilization is a vague and processor dependent concept, and you > > shouldn't expect to get any perfect answers. > > IMHO, one test on whether the change is an improvement is whether we're > providing data useful for answering higher level questions. > > For instance, for a workload I care about (building parts of solaris > from source), do I have make parallelism set optimally for this machine? > > mpstat on a multi-strand/hyperthreaded system isn't particularly useful > for that. > > - Bill >
Right. Talking about Saturation is a rathole better left to performance engineer. I think this is because something is saturated as soon as any sub-component of it is. So to evaluate the degree of saturation of X you need to consider every subsystem within it. In contrast 'utilisation' (or is it 'usage') is easily grasped and meaningful. It refers to the ratio of time a component was 'not idle' (thanks Bob for suggesting renaming %busy with %not idle). So when a system is idle we know we're leaving resources on the table. When it's 100% busy, it takes an expert to decide how much headroom, if any, there is. We won't cut the expert out with this exercise but we can help administrators out. So a CPU or Strand can be idle or busy and that is interesting. And a core can be idle or busy and that is interesting also. -r > > > > _______________________________________________ > perf-discuss mailing list > perf-discuss@opensolaris.org _______________________________________________ perf-discuss mailing list perf-discuss@opensolaris.org
