|
|
Log in / Subscribe / Register

Another push for sched_ext

Another push for sched_ext

Posted May 9, 2024 15:36 UTC (Thu) by flussence (guest, #85566)
Parent article: Another push for sched_ext

> > I fundamentally believe the approach to be detrimental to the scheduler eco-system. Witness the metric ton of toy schedulers written for it, that's all effort not put into improving the existing code.

No, no. There isn't a “scheduler ecosystem”. There is a scheduler *monoculture*, and for twenty years now people have interpreted that as damage and routed around it. The few of those who've dared to negotiate with the scheduler tyrant directly in the past have burned out and left kernel development forever - which is why it took a witheringly embarrassing fifteen years to properly address the Wasted Cores paper with more than a band-aid.

The only two choices remaining here are either accept the reality of the situation and address the root cause (in old hacker parlance, "maintainer needs face time with a LART"), or continue to be obstinate about it and encourage downstream hacks to proliferate - both individual and corporate. The corporations don't need royal consent to F up the kernels they ship on their devices, they've *been* doing it. And because they're butchering their device-specific kernels with hardwired hacks, for want of a sane pluggable mechanism, *users* are the only ones that get screwed because they can't turn it off.


to post comments

Another push for sched_ext

Posted May 9, 2024 15:48 UTC (Thu) by intelfx (subscriber, #130118) [Link] (2 responses)

> in old hacker parlance, "maintainer needs face time with a LART"

Could you please translate that? I’m not old hacker enough and that doesn’t seem to parse (or google).

Another push for sched_ext

Posted May 9, 2024 16:03 UTC (Thu) by schessman (subscriber, #82966) [Link]

LART : Luser Attitude Readjustment Tool. See Clue-by-four.

Another push for sched_ext

Posted May 9, 2024 19:56 UTC (Thu) by mathstuf (subscriber, #69389) [Link]

> that doesn’t seem to parse (or google).

FWIW, DDG gave the "Luser" expansion in its info box. My Google result has Wikipedia/Wiktionary as the first result (same thing).

Another push for sched_ext

Posted May 15, 2024 21:05 UTC (Wed) by Conan_Kudo (subscriber, #103240) [Link] (2 responses)

I personally would love to see sched_ext to land upstream for the simple reason that nobody upstream cares about desktop Linux and regular user needs for scheduling.

Having a way to build optimized schedulers outside of the kernel to bypass CFS and EEVDF and their flawed behaviors for regular users gives us an actual chance to deal with long-running responsiveness and performance-under-load issues that desktop Linux users observe that otherwise drive them to use alternative kernels that patch in other schedulers (like Con Kolivas' MuQSS and its predecessors).

Another push for sched_ext

Posted May 15, 2024 22:06 UTC (Wed) by jordalgo (guest, #170580) [Link]

Totally agree. It would be great to be able to experiment with new schedulers quickly.

Another push for sched_ext

Posted May 16, 2024 17:00 UTC (Thu) by Manifault (guest, #155796) [Link]

Improving the desktop experience was definitely one of the biggest targets we had in mind for potential improvements that could result from sched_ext. See this for an example of efforts to improve interactivity: https://drive.google.com/file/d/1fyHt9BYGha6apl7HAkibwpy5.... The techniques used there could also be added to EEVDF, which I'd be super happy to see.


Copyright © 2026, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds