Chris's Wiki :: blog/linux/MemoryRlimits Commentshttps://utcc.utoronto.ca/~cks/space/blog/linux/MemoryRlimits?atomcommentsDWiki2007-09-15T03:20:05ZRecent comments in Chris's Wiki :: blog/linux/MemoryRlimits.By Chris Siebenmann on /blog/linux/MemoryRlimitstag:CSpace:blog/linux/MemoryRlimits:979da231a6f71df3ab4b60d9409d539b97de082eChris Siebenmann<div class="wikitext"><p>Given the discussion of the expense of calculating PSS and USS in the
kernel, it doesn't seem like a real (in-kernel) PSS or USS limit would
be feasible or supported by the kernel developers.</p>
<p>The problem I see with a daemon is reacting fast enough to be useful
in our environment. In a compute server environment like what we're
running, I expect that most of the bad processes go bad very fast,
not grow slowly over a long time. (They may even start out bad
because they are simply being asked to work with a dataset that is
too big, so they OOM the machine as they try to start up.)</p>
<p>Unfortunately gathering actual information about this is hard, since
the kernel doesn't log enough of the right sort of information.</p>
</div>2007-09-15T03:20:05ZFrom 65.172.155.230 on /blog/linux/MemoryRlimitstag:CSpace:blog/linux/MemoryRlimits:ac43afa9feacfeb262137dd63f0504d27564c247From 65.172.155.230<div class="wikitext"><p>So I think you want is a PSS limit from: <a href="http://lwn.net/Articles/230975/">http://lwn.net/Articles/230975/</a> ... which, if someone gets PSS in, should be easier than RSS ... maybe. :).</p>
<p>Generally everyone seems to just go for ulimit -v, and not care about the mmap() large file users ... but I'm guessing you know that.</p>
<p>Another thing I'd been thinking about, which might help you, is having some kind of "nice daemon" ... this could monitor arbitrary conditions of processes on a box and then renice/iorenice/taskset or even SIGSTOP them for short amounts of time.
I have more than a slight suspicion that the only reason noone has written one already is due to the lack of a decent/fact API for reading /proc values.
Does that sound interesting?</p>
</div>2007-09-14T05:45:15Z