Marc-Adrian Napoli wrote:
>
> Dave,
>
> > > > > cache_mem 500 MB
> > > > >
> > > > The normal reccomendation is to leave this at the
> > > > default 8MB; what's your reasoning in using such
> > > > a large value?
> > >
> > > 1GB of RAM.
> > >
> > > Is that a good enough reason? (honestly).
> >
> > Nope. Squid's memory cache is probably not as efficient as your
> > filesystem's buffer-cache. For linux boxen, at least, I've narrowed the
> > optimal size to between 5 and 15MB. Got to find time to narrow it
> > further, if possible, and to check a few more kernels.
>
> So if squid only uses 5-15MB, the rest of the memory is used by linux
> correct?
<KOSH>
Yes
</KOSH>
(Except that, of course, the figure we are referring to is only a _part_
of squid's memory resources. It's the size of _one_ of squid's in-memory
structures. There are quite a few others)
> How can I stop linux from them swapping out? I've heard that disk swapping
> is a big no no on a box running squid... so I assumed that if you assigned
> all the memory to squid, it wouldn't attempt to swap out on itself.
>
> Hrm, i'm confused. :)
Under linux there's not really such a thing as idle memory. It'll get
used somewhere. That winds up being in disk-buffers, which is
more-or-less where we want it. Left for the kernel's use, it'll put it
to use, expand or reduce it as need be. Allocated to squid,
cache_mem+executable+index_size+transfer_buffers+a mess of other stuff
is the _least_ memory you will use. If that gets too big, you'll swap
and maybe die.
D
Received on Wed Nov 10 1999 - 20:11:58 MST
This archive was generated by hypermail pre-2.1.9 : Wed Apr 09 2008 - 11:57:32 MDT