Re: 2.6.16-rc5-mm1 -- strange load balancing problems

From: Peter Williams
Date: Thu Mar 02 2006 - 17:21:22 EST


Peter Williams wrote:
I'm seeing some strange load balancing problems with this kernel. I don't think that they're due to the smpnice patches as I've applied them on a standard 2.6.15-rc5 kernel and the problem doesn't happen there.

The problem is (as I say) quite strange and (for me) very reproducible. I have two programs (aspin and gsmiley) which I use to produce CPU hard spinners for testing purposes. What I'm finding is that when I start several copies of aspin load balancing goes as expected but when I launch several copies of gsmiley they all go to the one CPU and stick there like glue. (The most obvious difference between the two programs is that aspin is just a command line tool while gsmiley is an X windows program that spins a simley face and reports its own assessment of the percentage of CPU it's getting.) The machine that I've seen this problem is a hyper threading Pentium 4 and I suspect that it may be due to the SCHED_MC changes which overlap SCHED_SMT a bit.

I'm trying to test this on a non hyper threading machine but the machine has crashed (different kernel) while doing the build. I'll resume this effort tomorrow but I thought that I should report the problem so that others could comment.

Peter
PS SCHED_MC was configured in but I'll try it without tomorrow and report the results.

Configuring SCHED_MC to "no" causes this problem to go away.

Peter
--
Peter Williams pwil3058@xxxxxxxxxxxxxx

"Learning, n. The kind of ignorance distinguishing the studious."
-- Ambrose Bierce
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/