Re: [squid-users] Page cannot be displayed when cache fills up

26-Mar-02 at 12:39, warren, anthony (anthony.warren@medway.gov.uk) wrote :
> We are running 2.3STABLE4 on Redhat 7.1 (about to upgrade to 7.2).
>
> When the cache directory fills up, the users receive Page cannot be
> displayed error messages intermittently.
> I have attached our conifg to see if someone can spot something wrong with
> it.
> cache_mem 256 MB
> cache_dir ufs /opt/cache 43000 16 256

This is a commonly asked question. You have ~43Gb of cache, with only
256Mb of memory for it.

The "rule of thumb" is to have a 1:100 ratio of memory to cache, IIRC, so
you need at least, keeping round figures, 512Mb allocated.

That said, you will need more physical memory than that which you allocate
to cache_mem... probably at least twice as much.

So, your choices are: upgrade to more RAM or reduce Cache disk space.

I don't know what the current thinking is now, since disk space is so much
cheaper, but on graphs I saw, anything more than ~4Gb of cache_dir will
result in minimal improvement in actual hit rates, dependent of course on
how many users you have on the cache. 43Gb is a lot, though... so perhaps
reducing this to much less (go for 20Gb) will still give you very
reasonable hit rates / bandwidth savings.

-- 
[Simon White. vim/mutt. simon@mtds.com. GIMPS:56.19% see www.mersenne.org]
Microsoft isn't the answer. 
Microsoft is the question, and the answer is no.
[Arbitrary quotes signature rotation, a simple bash script by Simon White]