LWN: Comments on "Solving starvation problems in the scheduler"

Solving starvation problems in the scheduler

jospoortvliet — Thu, 20 Apr 2006 21:02:40 +0000

well, there simply ARE problems with the kernel scheduler... tough he
might have more problems with the I/O scheduler (he should try CFQ...), as
i said some time ago (below here), the cpu scheduler needs some work, too.
and imho id needs removal... and replacement.

Staircase scheduler

cyrus — Fri, 31 Mar 2006 01:09:46 +0000

Oh.. good to know, sorry for my previous post then, my fault.

Staircase scheduler

corbet — Fri, 31 Mar 2006 00:27:11 +0000

No, actually, the staircase scheduler is a CPU scheduler. See this LWN article from 2004.

Solving starvation problems in the scheduler

cyrus — Fri, 31 Mar 2006 00:06:14 +0000

What are you talking about? This article refers to the CPU SCHEDULER, which decides when which process is about to run. The Staircase scheduler from Con is an I/O scheduler, which controls a block device. (When to access disk and stuff..)

Solving starvation problems in the scheduler

jospoortvliet — Sun, 26 Mar 2006 20:07:17 +0000

as someone above said, this scheduler often comes up kind'a badly in the
news. really feels like a big hack... if you compare the design with dr.
Con Kolivas' [1] staircase scheduler - i really feel the kernel developers
should think about merging it.

its designed with interactivity in mind, not hacked on. all this fiddling
to work around a (seemingly) wrong design - it doesn't sound like the
right thing to do... and if it saves hundreds of lines of code, and
generally seems to perform better [2], wouldn't it be smarter to get it
in, even as an alternative choice, for even wider testing (altough it is
quite mature, after all these years of hard work)?

[1] http://members.optusnet.com.au/ckolivas/kernel/
[2] http://bhhdoa.org.au/pipermail/ck/2006-March/005693.html

Solving starvation problems in the scheduler

rahulsundaram — Fri, 24 Mar 2006 21:09:16 +0000

Could it be possible that you are running into a distribution specific issue rather than a kernel problem?

Solving starvation problems in the scheduler

xorbe — Fri, 24 Mar 2006 00:34:17 +0000

I apologize, but the whole Linux scheduler strikes me as so messed up. I have a 2.5GHz A64 w/2GB DDR500 that posts 7.8GB/s bandwidth in RMMA benchmark, with a recent hdd running 64-bit Linux -- and silly things like "ls" in the Mandriva cooker folder or copying to a USB2 drive bring the machine to a halt otherwise. My WinXP partition provides better interactivity, and it's starting to burn. I don't know what the problem is in the 2.6 series, but I wish it would get resolved soon.