EEVDF Scheduler Patches Updated For The Linux Kernel

[phoronix]

Phoronix: EEVDF Scheduler Patches Updated For The Linux Kernel

Intel Linux engineer Peter Zijlstra has sent out updated patches on the kernel patch series he's been working on the past several months around the "EEVDF" CPU scheduler for improving upon the current CFS scheduler code...

https://www.phoronix.com/news/Linux-EEVDF-EO-March

[ptr1337]

Really great work so far. Its not perfect and needs some more time, but it runs great.

Running the first posted version since 3 weeks on linux-cachyos and benchmarks are looking good. Also the responsives of the desktop feels better.
The second posted version does mostly fully remove CFS stuff. Personally I have a little regression in hackbench, when using hackbench -pTl . Default hackbench test seems to be better.

"perf sched ms fork thread" shows to be 50% better compared to CFS, which is great. Throughput is equal as it is with CFS in my benchmark suit.
Im looking forward to get it merged and the last regressions fixed.

Users from "linux-cachyos" are also quite happy with it.

[Quackdoc]

looking forward to trying this, but I dont feel like compiling kernels lately, and running arch. maybe I could pull from chaotic or cachyos, but I would rather not risk messing up my system again doing silly things. ill wait to see how it goes

[polarathene]

Is this meant to be similar to BFQ is for disks? (which made quite the difference in 4.x kernels before multi-queue, at least with slower disk I/O and low queues than what we have today).

[Mitch]

Really great work so far. Its not perfect and needs some more time, but it runs great.

Running the first posted version since 3 weeks on linux-cachyos and benchmarks are looking good. Also the responsives of the desktop feels better.
The second posted version does mostly fully remove CFS stuff. Personally I have a little regression in hackbench, when using hackbench -pTl . Default hackbench test seems to be better.

"perf sched ms fork thread" shows to be 50% better compared to CFS, which is great. Throughput is equal as it is with CFS in my benchmark suit.
Im looking forward to get it merged and the last regressions fixed.

Users from "linux-cachyos" are also quite happy with it.

If I'm reading the patch and your comments correctly, this scheduler appears to have the same bandwidth as CFS, and better latency across the board. If so, I'd love if they could merge this work as soon as possible somehow.

[Linuxxx]

If I'm reading the patch and your comments correctly, this scheduler appears to have the same bandwidth as CFS, and better latency across the board. If so, I'd love if they could merge this work as soon as possible somehow.

True, if it can really keep the great throughput performance of CFS while still being able to reduce the overall latency, then it would be simply outstanding!

Although I must admit that a properly configured Linux kernel can already provide a smooth experience, such as Ubuntu's "lowlatency" flavor.

[geearf]

Really great work so far. Its not perfect and needs some more time, but it runs great.

Running the first posted version since 3 weeks on linux-cachyos and benchmarks are looking good. Also the responsives of the desktop feels better.
The second posted version does mostly fully remove CFS stuff. Personally I have a little regression in hackbench, when using hackbench -pTl . Default hackbench test seems to be better.

"perf sched ms fork thread" shows to be 50% better compared to CFS, which is great. Throughput is equal as it is with CFS in my benchmark suit.
Im looking forward to get it merged and the last regressions fixed.

Users from "linux-cachyos" are also quite happy with it.

How does it compare to BMQ/PDS?

[ptr1337]

How does it compare to BMQ/PDS?

I don't use BMQ/PDS. It is currently full of bugs and should be avoided until these bugs are resolved or the scheduler gets rewritten.

[geearf]

I don't use BMQ/PDS. It is currently full of bugs and should be avoided until these bugs are resolved or the scheduler gets rewritten.

Oh, what are the bugs? I've been using them since TKQ recommended them but I don't do much on my PC anymore so I could miss stuff. Thanks!

[ptr1337]

Oh, what are the bugs? I've been using them since TKQ recommended them but I don't do much on my PC anymore so I could miss stuff. Thanks!

Just follow the issues here:

https://gitlab.com/alfredchen/linux-prjc/-/issues/

TKG does also not really suggest pds/bmq anymore much due the recent issues with newer kernels.
CFS/BORE/EEVDF are the way to go. CFS improved a lot.

I just saw there was some recent activity at prjc, but the root issues are still open at all.