RE: [squid-users] Disks configuration

> What I've done so far is:
> Brought the machines to 1GB RAM
> Reset cache_mem and cache_dir as follows:
> cache_mem 300

That's 300 Mb, I hope.

> cache_dir ufs /usr/local/squid/cache 45000 128 256
>
> Obviously, the "cache-size" had to be dropped or the filesystem would
> have eventually flooded. Secondly, with the large cache size, I felt
> it much better to have a "broader" tree allowing for smaller
> individual
> directories.
>
> My other thought is to break up the stripe and specifically
> define three
> separate cache_dirs. The systems owners have also asked if I'd be
> willing to upgrade to 2.4STABLE1 (obviously, the answer there is yes).

> I'd like to know the list members' opinions on what I've done
> and on the
> proposed unstriping.

Obviously you did the Right Thing(tm)

If I were you, I'd recommend to unstripe, and to drop ufs in favour of some
asynchronouns filestorage (diskd or aufs). As it is now, you can only
have one outstanding I/O request at a time, and until it completes squid is
blocked.
Having more cache_dirs will improve parallelization (maybe I overdid it, but
I have two cache_dirs per spindle to try and exploit buffer elevator
algorithms
in fast file accesses. How much this helps - if it does at all - is up to
debate though).

-- 
	/kinkie