Re: High priority tasks break SMP balancer?

On Fri, Nov 16, 2007 at 07:07:00AM +0100, Ingo Molnar wrote:

* Micah Dowty <micah@xxxxxxxxxx> wrote:

I am a bit at a loss as to how this could relate to the patch. This
looks like a load balance logic issue that causes the load
calculation to go wrong?

My best guess is that this has something to do with the timing with
which we sample the CPU's instantaneous load when calculating the load
averages.. but I still understand only the basics of the scheduler and
SMP balancer. All I really know for sure at this point regarding your
patch is that git-bisect found it for me.

hm, your code uses timeouts for this, right? The CPU load average that
is used for SMP load balancing is sampled from the scheduler tick - and
has been sampled from the scheduler tick for eons. v2.6.23 defaulted to
a different method but v2.6.24 samples it from the tick again. So my
guess is, your testcode behave similarly on 2.6.22 too, correct?

Interesting.. here are the kernels I've tested so far, not including
the git-bisect run I did between 2.6.19 and 2.6.20:

2.6.17 -
2.6.19 -
2.6.19.7 -
2.6.20 +
2.6.21 +
2.6.22 -
2.6.23.1 +

Here a "-" means that the problem does not occur (my test program uses
100% of both CPUs) and a "+" means that the test program leaves one
CPU mostly idle.

Unless I've made a mistake, 2.6.22 seems like the outlier rather than
2.6.23. Is this inconsistent with the scheduler tick hypothesis?

Thanks,
--Micah
-
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/

Relevant Pages

  • Re: [RFC][PATCH 0/6] Add group fairness to CFS - v1
    ... new precise stats code there is now default-enabled - making SMP ... So where's this precise stats based calculation of cpu_load? ... now we basically use a load derived from the fair-time passed ... value that was sampled by the scheduler tick. ...
    (Linux-Kernel)
  • Re: VST and Sched Load Balance
    ... since the idle CPUs will no longer pull tasks from ... I am looking at the recent changes in load balance and I see that load ... it can decide to not pull tasks. ...
    (Linux-Kernel)
  • Re: Multiple Terminal Servers Question
    ... You can set it to load balance ... will get all load information off the TSs (like CPU, Memory, Number of Users ... I have set up Terminal Services Roaming profiles in AD. ... > another server I can. ...
    (microsoft.public.windows.terminal_services)
  • Re: Oracle 10g TAF Vs FAN
    ... Obvious reason for the load is ... But my doubt is, if the load balance is enabled or working fine, during the ... To make this work load evenly distributed between 2 nodes, what configuration ... " I have 2 node cluster and as of now only 1 of the node is having the full load." ...
    (comp.databases.oracle.server)
  • Re: High priority tasks break SMP balancer?
    ... looks like a load balance logic issue that causes the load ... calculation to go wrong? ... has been sampled from the scheduler tick for eons. ...
    (Linux-Kernel)