What scheduler produces the best results on regular desktop systems? mar., 2 mai 2023, 09:54 Holger Hoffstätte <hol...@applied-asynchrony.com> a scris:
> On 2023-05-01 22:24, Nikos Chantziaras wrote: > > On 26/04/2023 23:06, Holger Hoffstätte wrote: > >> On 2023-04-26 18:15, Nikos Chantziaras wrote: > >>> So I wanted to try the BMQ scheduler in gentoo-sources. Is just > >>> enabling it in the kernel build all that's needed? I did so, > >>> booted, and: > >>> > >>> $ dmesg | grep -i bmq [ 0.100284] sched/bmq: BMQ CPU Scheduler > >>> v6.1-r4 by Alfred Chen. > >>> > >>> That's all and it's in use now? Or do I need to toggle anything > >>> else? > >> > >> That's all you need to do, right. You can slso alternatively switch > >> to PDS under the "General setup > Scheduler features" kernel-config > >> menu, which is slightly better for throughput and "more correct" > >> when many tasks have varying priorities - beefy workstation or > >> server. BMQ is kind of simplistic (in an elegant way), but that's > >> why it is so effective for low-end systems and desktops. Run iperf3 > >> over loopback (i.e. both server and client) to see the difference. > > > > Switched back to the default (called "CFS" I think.) > > > > BMQ has severe issues. When emerging something while I play a game > > (either native or through wine-proton,) there's long lag spikes and > > freezes. Even worse, there's bugs like the system completely hanging > > on shutdown, or "umount" hanging with 100% CPU use by two kernel > > threads. > > Can you try booting with psi=0 and check if that helps? There's a known > bad interaction with the PSI (Pressure Stall Information) mechanism, > which is active by default in many kernel configs. The latest version > of the BMQ patch makes sure to turn it off, you may not have that version > yet. BMQ (and PDS) works fine and has done so for quite a while. > > -h > >
