BFS Scheduler Gets "Substantial Updates" For Linux 4.7

[phoronix]

Phoronix: BFS Scheduler Gets "Substantial Updates" For Linux 4.7

Con Kolivas announced this week BFS 497, a major new release of his scheduler that's now fitted for the Linux 4.7 kernel...

http://www.phoronix.com/scan.php?pag...uler-Linux-4.7

[SaucyJack]

Awesome, it's a damn shame i need bleeding edge kernels to get the amd improvements though

[meekteef]

I hope it gets added to the PF kernel on Fedora soon.

[Zan Lynx]

Does it still lose out on efficiency vs CFS on 6+ core machines?

[cl333r]

Does it still lose out on efficiency vs CFS on 6+ core machines?

Up to 16 cores is fine [1] - source from 2011, since then BFS likely improved further.

[1]
http://ck-hack.blogspot.md/2011/04/s...ty-of-bfs.html

[alexhaydock]

I notice the wording in the notes for the patch file states:

Scalability is optimal when your workload is equal to the number of CPUs on
bfs. ie you should ONLY do make -j4 on quad core, -j2 on dual core and so on.

Does anyone know how hyperthreading interacts with this advice? I'm guessing it's still optimal that for a quad core with hyperthreading, for instance, to be using -j8 rather than -j4?

[cj.wijtmans]

I notice the wording in the notes for the patch file states:

Does anyone know how hyperthreading interacts with this advice? I'm guessing it's still optimal that for a quad core with hyperthreading, for instance, to be using -j8 rather than -j4?

yes you use -j9 on hyperthreaded quad core. And in fact you use 1+cores not equal to cores.

[Vash63]

Have there been any benchmarks in the last few months for the ck patchset? Wondering if it still makes a difference with all of the improvements the vanilla kernel has had lately.

[Uqbar]

The real shame is that this tiny piece of computer programming art is kept off the official kernel options.

Linux should all be about options. But only those liked by Linus.

I wish Con dare to push it again for inclusion.

[VikingGe]

Up to 16 cores is fine [1] - source from 2011, since then BFS likely improved further.

The problem is that CFS has improved even more in my experience. I've got a 6-core machine and single-threaded tasks run at like half the speed on BFS compared to CFS because of massive core hopping issues on BFS. Haven't tested the newest update, though.