Messages in this thread | | | Date | Tue, 27 Sep 2011 15:39:09 +0200 | From | Ingo Molnar <> | Subject | Re: [PATCH v1] sched: fix nohz idle load balancer issues |
| |
* Srivatsa Vaddagiri <vatsa@linux.vnet.ibm.com> wrote:
> * Ingo Molnar <mingo@elte.hu> [2011-09-27 08:32:24]: > > > What are the tasks doing which are running - are they plain burning > > CPU time? If the tasks do something more complex, do you also have a > > measure of how much work gets done by the workload, per second? > > They are simple cpu hogs at this time. > > > Percentual changes in that metric would be nice to include in an > > additional column - that way we can see that it's not only idle > > that has gone down, but workload performance has gone up too. > > Ok, good point. > > > In fact even if there was only a CPU burning loop in the workload it > > would be nice to make that somewhat more sophisticated by letting it > > process some larger array that has a cache footprint. This mimics > > real workloads that don't just spin burning CPU time but do real data > > processing. > > > > For any non-trivial workload it's possible to reduce idle time > > without much increase in work done and in fact it's possible to > > decrease idle time *and* work done - so we need to see more clearly > > here and make sure it's all an improvement. > > Ok - I will run a cpu intensive benchmark and get some numbers on > how benchmark score varies with the patch applied. I can pick a > simple matrix multiplication type benchmark, unless you have other > suggestions!
Yeah, matrix multiplication would be fine i think.
You could create it yourself and add it into tools/perf/bench/cpu-matrix.c as a new 'perf bench cpu matrix' testcase - perhaps with a '-s 10m' parameter that defines the size of the matrices, and a '-i 1000' parameter to specify the number of iterations - or so.
We could use that for scheduler HPC benchmarking in the future as well.
Thanks,
Ingo
| |