LWN: Comments on "CFS group scheduling"

CFS group scheduling

kolyshkin — Tue, 28 Aug 2007 15:41:18 +0000

This type of scheduling is called "group scheduling"; Linux has never really supported it with any scheduler.

Well, whe scheduler we have in OpenVZ is doing just that. It's a two-level CPU scheduler. On the first level schedules between groups of processes (with each group being a container), taking into account the (relative) priorities for those groups, and the (absolute) limits on the CPU time being used. On the second level, it schedules the process within chosen group.

That's indeed nice that such a feature appears in vanilla kernel.

Userspace scheduling policy daemon

TRS-80 — Fri, 06 Jul 2007 10:11:46 +0000

The usual way to handle this is child processes get put into the same group as their parent. Obviously you wouldn't want this for all processes, but the daemon could mark things like make and gcc as creating a new sub-hierarchy.

CFS group scheduling

davecb — Thu, 05 Jul 2007 14:35:36 +0000

I've been a happy user of this kind of functionality
in Solaris 9 and 10 (before that it was pretty primitive).

I originally used for production machines, where a
guarantee of a certain amount of CPU to a bunch of
processes allows for eevrything from consolidation
to having the CPU available for root to use to kill
a runaway process.

However, I now run it on my laptop, and shove
background proceses into a different scheduling
group so they don't interfere with anything in
the foreground. Think of this as bg and nice
done right (;-))

--dave

Userspace scheduling policy daemon

abatters — Thu, 05 Jul 2007 13:49:11 +0000

> So any particular policy can be implemented through the creation of a simple, user-space daemon which responds to process creation events by placing newly-created processes in the right group.

Using a userspace daemon to set process scheduling policy might be fine for long-running processes like the X server, but wouldn't it introduce a lot of overhead for short-lived processes like gcc during kernel builds? I expect it would add a few context switches of overhead to every fork(); that doesn't seem consistent with general kernel developer attitude towards efficiency.

CFS group scheduling

mclasen@redhat.com — Thu, 05 Jul 2007 13:33:55 +0000

Might be worthwhile to point out that the CFS scheduler has been included in the Fedora devel kernel for a few weeks now.