Re: Scheduler(?) regression from 2.6.22 to 2.6.24 for short-lived threads

[Olof Johansson <olof@lixom.net>]

On Mon, Feb 11, 2008 at 09:15:55AM +0100, Mike Galbraith wrote:
> Piddling around with your testcase, it still looks to me like things
> improved considerably in latest greatest git.  Hopefully that means
> happiness is in the pipe for the real workload... synthetic load is
> definitely happier here as burst is shortened.

The real workload doesn't see much of an improvement. The changes I did
when tinkering yesterday seem like they're better at modelling just
what's going on with that one.

I've added the new testcase I'm using for the numbers I posted last
night at http://lixom.net/~olof/threadtest/new/, including numbers for
the various kernels. I also included the binaries I built. (with "gcc
-DLOOPS=<size> testcase.c -lpthread").

I tried graphing the numbers as well, it looks like for larger workloads
that 2.6.22 has a fixed benefit over newer kernels. I.e. it seems quicker
at rebalancing, but once things balance out they're obviously doing ok
independent of kernel version.

Graph at http://lixom.net/olof/threadtest/new/schedgraph.pdf. I couldn't
figure out how to make the X axis linear, so it obviously looks odd with
the current powers-of-two at the end instead of linear, but the
differences can still be seen clearly.


-Olof