[squid-users] Random dropping of requests

Greetings,

I have just taken charge of a squid machine that has been showing some
strange behaviour for the last few weeks. For one thing, cpu usage on
average is above 80% which seems abnormally high, even taking into
account the high traffic (about 2000 hits per minute).

Secondly, all traffic randomly stops for anywhere between a few seconds
and 20 minutes. Restarting squid during this time brings it back up, but
this isn't feasable when we have to do it a few times a day. Even if I
don't restart manually, it becomes available again by itself, but the
downtime is too costly.

Now for some details. The hardware is as follows:

2.7GHz x 2 Xeon CPUs
2GB RAM
36GB x 2, 10,000 rpm SCSI drives

One drive is used for the OS (Fedora Core 2, default kernel) while the
other is used for cache (32GB cache size, reiserfs with notail and
notime options).

Next, until a few days ago, the machine was running Squid 2.5-STABLE6
(without patches). It was configured to use diskd and various options
which I don't have availabe right at this minute.

Quite a few OS settings recommended for Squid are present, such as port
range (1024-65000), file descriptors (32768), tcp_max_syn_backlog (8192)
etc.

I tried switching to 2.5-STABLE5 with the 3 major patches since we have
this setup on another proxy, but got the same results. This morning, I
upgraded to the latest 2.5-STABLE6 version with all the current patches
and set the cache type to aufs (simple 'squid -z', not a reformat of the
underlying filesystem), but it solved neither the CPU usage issue, nor
the random drops.

Nothing unusual shows up in cache_log, access_log or any of the system
logs. Our network monitor also confirms that this isn't a network issue
and that requests are going to the proxy during the outages.

Unless someone can identify the problem here, I'll consider installing
Slackware and a custom kernel to rule out that it's an OS issue.

Any help will be appreciated.

Thanks,

-- 
A. Sajjad Zaidi
GnuPG Key ID: 0xD7AD0E13