On 06/19/20 12:57, Mel Gorman wrote: > On Fri, Jun 19, 2020 at 11:36:46AM +0100, Valentin Schneider wrote: > > > nouclamp uclamp > > > uclamp-static-key > > > Hmean send-64 162.43 ( 0.00%) 157.84 * -2.82%* > > > 163.39 * 0.59%* > > > Hmean send-128 324.71 ( 0.00%) 314.78 * -3.06%* > > > 326.18 * 0.45%* > > > Hmean send-256 641.55 ( 0.00%) 628.67 * -2.01%* > > > 648.12 * 1.02%* > > > Hmean send-1024 2525.28 ( 0.00%) 2448.26 * -3.05%* > > > 2543.73 * 0.73%* > > > Hmean send-2048 4836.14 ( 0.00%) 4712.08 * -2.57%* > > > 4867.69 * 0.65%* > > > Hmean send-3312 7540.83 ( 0.00%) 7425.45 * -1.53%* > > > 7621.06 * 1.06%* > > > Hmean send-4096 9124.53 ( 0.00%) 8948.82 * -1.93%* > > > 9276.25 * 1.66%* > > > Hmean send-8192 15589.67 ( 0.00%) 15486.35 * -0.66%* > > > 15819.98 * 1.48%* > > > Hmean send-16384 26386.47 ( 0.00%) 25752.25 * -2.40%* > > > 26773.74 * 1.47%* > > > > > > > Am I reading this correctly in that compiling in uclamp but having the > > static key enabled gives a slight improvement compared to not compiling in > > uclamp? I suppose the important bit is that we're not seeing regressions > > anymore, but still. > > > > I haven't reviewed the series in depth because from your review, another > version is likely in the works. However, it is not that unusual to > see small fluctuations like this that are counter-intuitive. The report > indicates the difference is likely outside of the noise with * around the > percentage difference instead of () but it could be small boot-to-boot > variance, differences in code layout, slight differences in slab usage > patterns etc. The definitive evidence that uclamp overhead is no there > is whether the uclamp functions show up in annotated profiles or not.
I certainly have seen weird variations in the numbers. If you've seen my numbers in the links below, I was buffled when I moved to 5.7-rc2 and couldn't reproduce again. https://lore.kernel.org/lkml/20200616110824.dgkkbyapn3io6wik@e107158-lin/ I think the hot path can be sensitive to code/data layout variations and now uclamp added more variables to be accesses, this sensitivity could be manifested in more ways, me thinks. I am re-running the test now with perf record. But not sure if I'll be able to provide the numbers by the end of the day. If it is easy for you to pick this up, I'd appreciate if you can kick off a test. But it's Friday after all.. :-) Thanks -- Qais Yousef
