[systemd-devel] [PATCH] cgroups: Cache controller masks and optimize queues.

[Joe Miller]

Tested this branch with 16,000 units and 32,000 units and it looks great.
Much better than before.


On Mon, Nov 18, 2013 at 11:12 PM, David Timothy Strauss <
david at davidstrauss.net> wrote:

> I'm pretty confident in the accuracy of the controller mask
> aggregation, especially given the new unit test. Here are the main
> review questions.
>
> Is it okay to couple realizing the cgroups and applying the attributes
> the way the patch does? Before, we were often applying the attributes
> for all of a unit's siblings (read: other units in the same slice). I
> believe these should never change as long as the controller mask is
> the same as what's been realized already. Perhaps the old
> implementation was done to avoid re-computation of the target masks
> (which is no longer required).
>
> Is it okay to skip realization and attribute setting for
> inactive/failed units? If they do start, we will set the attributes at
> that time with this patch. If they're not running, giving them CPU and
> Block IO weights doesn't seem to have any worthwhile effect. Am I
> ignoring something about, say, CPU accounting (which I don't use
> myself)?
>
> Also, hat tip to my coworker Kyle Ibrahim for the original idea of not
> queuing inactive and failed units.
> _______________________________________________
> systemd-devel mailing list
> systemd-devel at lists.freedesktop.org
> http://lists.freedesktop.org/mailman/listinfo/systemd-devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/systemd-devel/attachments/20131121/48e8d927/attachment.html>