User: Password:
|
|
Subscribe / Log in / New account

Adding periods to SCHED_DEADLINE

Adding periods to SCHED_DEADLINE

Posted Jul 20, 2010 22:04 UTC (Tue) by xxiao (guest, #9631)
Parent article: Adding periods to SCHED_DEADLINE

how about a new scheduler option...CFS is hard to tune and has a really bad performance on my multithread embedded apps(>10% cpu usage increase,etc), the same app ran perfectly well with the O1 scheduler in the past for years.


(Log in to post comments)

Adding periods to SCHED_DEADLINE

Posted Jul 21, 2010 7:46 UTC (Wed) by abacus (guest, #49001) [Link]

Please report this as a kernel bug, such that Ingo Molnar can have a look at it.

Adding periods to SCHED_DEADLINE

Posted Jul 21, 2010 16:18 UTC (Wed) by cesarb (subscriber, #6266) [Link]

It would be cool if people could "record" a trace of their scheduling so that the sched maintainers could "replay" it, so that they can find and fix bugs without having to find a reproducible testcase (i.e. even if it happened just once, if you have a trace, it could be used to find the cause).

I do not know if this already exists; I have not looked.

Adding periods to SCHED_DEADLINE

Posted Jul 21, 2010 19:31 UTC (Wed) by sbohrer (guest, #61058) [Link]

I believe "perf sched record" and "perf sched replay" will do exactly that.

Adding periods to SCHED_DEADLINE

Posted Jul 21, 2010 8:01 UTC (Wed) by viiru (subscriber, #53129) [Link]

> how about a new scheduler option...CFS is hard to tune and has a really
> bad performance on my multithread embedded apps(>10% cpu usage
> increase,etc), the same app ran perfectly well with the O1 scheduler in
> the past for years.

Google is having the same problem with some of their apps, which is why they have ported the old scheduler over to the 2.6 series. I'm not sure if they have a kernel tree with that publicly available, though.


Copyright © 2017, Eklektix, Inc.
Comments and public postings are copyrighted by their creators.
Linux is a registered trademark of Linus Torvalds