Nice, thanks! :) Honestly, I have no idea why taskgetid benchmark, of all things, gets so much faster. I'm tempted to look into the cause now.
-- jzr On Tue, Jan 23, 2024, 1:36 PM Jiri Svoboda <[email protected]> wrote: > Hi, > > while I didn't have the time yet to go through the scheduler changes in > detail to try and understand them, I was wondering what (if any) effect > they have on performance. I ran a couple of benchmarks. Here's the results: > > Qemu + KVM amd64 > ================ > top busy 6.5-7.4% both > ping_pong old: 70167 ops/s new: 77737 ops/s > taskgetid: old 3637731 ops/s new 5833524 ops/s (!!!!) > read1k: old 13826 ops/s new 15073 ops/s > fibril_mutex: old 13432519 ops/s new 13497358 ops/s > malloc1: old 5878492 ops/s new 5870265 ops/s > > Qemu without KVM amd64 > ====================== > top busy: old 50% new: 50%c > ping_pong: old 2519 ops/s new 2697 ops/s > taskgetid: old 625269 ops/s new 1389736 ops/s (!!!!) > read1k: old 634 ops/s 656 ops/s > fibril_mutex: old 572367 ops/s 585143 ops/s > malloc1: old 355691 ops/s 345153 ops/s > > The most obvious is that taskgetid performance increased about 2x (!!!!) > We can also see a more moderate (10%) increase in ping_pong and still more > moderate in read1k. fibril_mutex and malloc1 are the same within the error > margin. > > Regards, > Jiri > _______________________________________________ > HelenOS-devel mailing list > [email protected] > http://lists.modry.cz/listinfo/helenos-devel >
_______________________________________________ HelenOS-devel mailing list [email protected] http://lists.modry.cz/listinfo/helenos-devel
