Re: [squid-users] AUFS number of threads

From: Ron Vachiyer <proutfoo@dont-contact.us>
Date: Tue, 02 Apr 2002 11:12:53 -0500

.
>
>How many drives do you have?
>
>General rule: Not more than 12 per drive and not less than 6 per drive,
>and probably not more than 50 in total.. The default should fine in most
>situations.
>
>As for exact rule, only experimentation can tell. It depends on your
>traffic pattern, number of drives, speed of drives and CPU. But as long
>as you have a reasonable amount you should be fine.
>
>To few, and you hit ratio will suffer (and there will be lots of
>complaints in cache.log)
>
>To many, and hit latency will suffer due to overly long disk I/O queues,
>possibly even causing Squid to spiral down..

Hello,

I am beginning to suspect AUFS as a cause to my silent signal-6 squid
aborts. I have 52 squid processes running on 3 disks with AUFS, and I have
not attempted to tune the --enable-async-io[=N_THREADS]. My question is,
the N_THREADS, is that total possible threads or number of threads
per-cache_dir?

What is strange also is since I have upgraded from a Redhat6.2 (with Linux
2.2.20) to a Redhat 7.2 (with kernel 2.4.9-31 rh proprietary), whenever I
rotate the logs, squid no longer processes requests until it rebuilds or
checks the store. Is it possible to rotate the logs without having squid
re-check its contents? My main goal in rotating the logs is to keep the
size of the logfiles down to manageable chunks. In order to do this, I must
rotate every two hours or so.

Thanks,

Ron

_________________________________________________________________
Send and receive Hotmail on your mobile device: http://mobile.msn.com
Received on Tue Apr 02 2002 - 09:13:54 MST

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 17:07:19 MST