Fwd: [PATCH] Add load average support to makehistory
Russ Allbery
rra at stanford.edu
Mon Jun 22 16:41:05 UTC 2009
"Jonathan Kamens" <jik at kamens.brookline.ma.us> writes:
> Russ Allbery wrote, "This seems like sort of a hacky way to
> approximate the same thing that nice would do in a much better and
> finer-grained way. But maybe I'm missing something?"
>
> The problem is that nice will not handle the problem the patch is
> trying to solve in a better and finer-grained way. In fact, nice
> won't help with the problem at all.
>
> Nice changes the scheduling priority of a process, i.e., what
> proportion of available CPU cycles it is allocated. However, the
> resource that gets overwhelmed when makehistory is running is not CPU,
> but rather disk I/O. When makehistory hoses the disk, a bunch of
> processes will get stuck in disk wait, which means that there will be
> plenty of CPU available for the processes that aren't, which means
> that even if you nice makehistory all the way up to 20, it'll still
> get plenty of CPU to continue hosing the disk.
Oh, huh. Wow, that feels like a bug in the way that prioritization is
handled to me, but I can see how it would fall out that way. In that
case, I have no objections to apply this patch.
Do you think this method is needed for anything other than makehistory?
Do you have trouble with expireover, for instance?
--
Russ Allbery (rra at stanford.edu) <http://www.eyrie.org/~eagle/>
Please send questions to the list rather than mailing me directly.
<http://www.eyrie.org/~eagle/faqs/questions.html> explains why.
More information about the inn-workers
mailing list