Re: [squid-users] squid slow

On Mon, 2009-08-24 at 18:30 +1200, Amos Jeffries wrote:
> Muhammad Sharfuddin wrote:
> > On Mon, 2009-08-24 at 17:05 +1200, Amos Jeffries wrote:
> >> On Mon, 24 Aug 2009 10:24:41 +0600, Muhammad Sharfuddin
> >> <m.sharfuddin_at_nds.com.pk> wrote:

> >
> > So I really dont understand why you said/wrote 'The following rules
> > *ONLY* apply to external people'
> >
>
> Because you "allow localnet" (AKA unrestricted access to all internal
> client) before doing those rules.
>

> Regardless of what they are they will only be tested against the
> requests coming from outside your "localnet" defined local network ranges.
thanks
in squid.conf
#http_access allow localnet
then
# squid -k reconfigure
>

> > you are recommending aufs over diskd, and the following url suggest 'diskd'
> > as the the store type of choice for the Cache-off's
> > http://www.linuxsa.org.au/pipermail/linuxsa/2004-June/070228.html
> >
>
> Written in 2004. Server CPU threading has come a long way since then.
>
> diskd is a single-threaded helper application, with a processing IO
> _upper_ limit of 1 file read at a time. Squid itself does not block, but
> the helper reads/writes are blocking each other _within the helper_.
>
> AUFS is a multi-threaded component utilizing the kernel and all
> available CPU for non-blocking read/write to as many files as needed
> simultaneously. Limits are defined by the available FD in Squid and
> system CPU capabilities.
>
> diskd is only recommended for use on *BSD systems where AUFS support is
> not available (yet).
>
Ok, nice and thanks
> >
> >> Also with ~50GB of storage you are probably wanting to use something like
> >> 32 or 64 for the Level-1 value (currently 16). Changing that requires a
> >> cache delete and rebuild with 'squid -z' though.
> > whats the rule/formula for Level-1 and Level-2 value ? is it related
> > with storage size ?
> >
>
> Yes. OS used to have an upper limit on the number of files stored in a
> single directory. I think most still do for the common filing systems.
>
> Between them these numbers define how many folders are used in the
> cache. Smaller caches only need a few folders, bigger caches need a lot
> more to keep the OS happy.
>
> The default squid.conf comes tuned for a 200MB cache. Quite small for
> any real use. When you are heading into tens of GB its a good idea to
> start upping these numbers. How much depends on your OS filesystem and
> avg. object size in the cache. Big and huge objects obviously reduce the
> pressure for extra folders.
>
cache_dir aufs /var/cache/squid 50000 32 256

so the above one is OK, or should I also need to increase the Level-2
value(256) ?
By the way whats the best file system for caching reiserfs, ext3, or XFS
( I am using reiserfs with 'notail,noatime,noacl' options)

> >>
> >> These days I'm advising people terminate their file extension patterns with
> >> (\?.*)?$ instead of just $ to catch all the sites using dynamic parts in
> >> their URLs.
> >>
> >>
> > you mean the following ?
> > (\?.swf)?$
> > (\?.mdi)?$
> > e.g
> > refresh_pattern -i (\?.swf)?$ 43200 100% 43200 override-lastmod
> > override-expire
>
> No, no.
>
> This:
> refresh_pattern -i \.swf(\?.*)?$ ....
Thanks

Regards
--ms
Received on Mon Aug 24 2009 - 06:48:22 MDT