Re: [squid-users] Single processor?

Thank you all for the excellent suggestions. I've installed a new
primary Squid server (which I'll be setting up a second, ICP machine to
split load on.)

It is a Dell Poweredge 2500, dual 1g P3 (?) machine with 6 16.9g scsi
drives and 1g RAM. It is still using the stock Redhat 7.1 (2.4.2smp2)
kernel which, as suggested, I'll be upgrading soon. I have a 2 disk
mirrored set for the system, squid bin's, and logs. 3 of the 16.9g
drives are cache dirs (caches 1, 2, and 3) with 3 seperate 'cache_dir'
entries in squid.conf (all as suggested--my boss and I found this
technique on the internet before I checked my mailing list logs.) The
machine is using 2 NICs--access to the internet is on one and the
intranet on the other (default gw). Usage seems to be good on this
machine, unlike the other one with the RAID set holding the cache.
 Worse than that, the other's RAID set has no write-back cache, so when
lots of writes got queued the machine would load up as the writes
bursted out--not good.

I appreciate the suggestions as they were all on the mark. I'll most
likely be upgrading the cache store to aufs as well, instead of the
default ufs.

Thanks!

Peter Smith

Colin Campbell wrote:

>Hi,
>
>On Tue, 25 Sep 2001, Billy Huddleston wrote:
>
>>What about RAID 0? Has anyone done in comparison vs multiple cache dirs?
>>
>
>It'll depend on a whole lot of things. How big is the stripe? How many
>cache objects fit in one stripe? If the stripe is large you are efectively
>getting lots of small cache_dirs. If the stripe is small, you'll need to
>access multiple disks to access objects. If one disk dies your whole cache
>is gone.
>
>Colin
>
Received on Thu Sep 27 2001 - 17:50:57 MDT