Hi,
I asked about this problem last week, I just thought I'd let people know
what has happened.
On Wed, 24 Sep 1997 11:39:39 +0100 (British Summer Time) Nick O'Brien
<nick@cant.ac.uk> wrote:
> I am running Squid v1.1.11 on a system running Digital Unix 3.2C with
> 74M of real memory, and 128M of swap space.
>
> This morning accessing pages through Squid was running very slowly. To
> give you some idea of how slow a particular page retrieved directly
> from our web server (which is on a different system) without a proxy
> or using our old CERN proxy (which is on the same system as Squid)
> takes 1-2 seconds to load. However with the Squid proxy it takes
> 6-7 seconds. Even after restarting Squid the problem continues.
One suggestion was to upgrade to the latest version of Squid, so I ugraded
to v1.1.16 - this had no effect on the situtation. As performance appeared
to better early in the morning and in the evenings someoner suggested that
I might be running out of file descriptors, but Cache Manager claimed that
I was only using 90 out of 4096.
Some suggested that I might be running out of file descriptors, but the
Cache Manager Cache Information Report claimed that I was only using about
90 out of 4096 file descriptors.
I came in on Saturday and Squid was as slow as ever, so I was wrong - it
is not related to the number of users accessing Squid. Following on from
an idea someone's email had given me I reduced the value of cache_swap by
7/8 (3950 to 500). I then restarted Squid - this had no effect. But
when I removed the the Cache swap log file (cache_swap_log) and restarted
Squid - performance picked up instantly.
Therefore what I suspect is happening is that Squid is spending so much
time checking to see if items are in the Cache swap log file, that it has
little time to do anything else - including retrieving items directly from
source. If this is the case I obviously want to find a way of speeding up
access to the log file as I don't want to have to destroy my cache every
few days to keep performance at a good level (access is reasonable at the
moment). I've also reduced the value of reference_age to 1 month to see if
this will help.
Do you have any ideas of what might be wrong?
Also - I now unable to use cachemgr.cgi as I had to switch off our Unix
based web server. Is there an NT Alpha binary of cachemgr.cgi (I don't
have any access to any NT based compilation tools so I can't recompile the
source), or at the very least a Perl version available?
Rgds.,
Nick.
# Nick O'Brien, Computer Officer "It gives me a headache just to #
# Canterbury Christ Church College think down to your level", Marvin #
# Phone: +44 1227 782468 the Paranoid Android, HHGTTG #
# Email:nick@cant.ac.uk Web:http://www.cant.ac.uk/staff/nick/home.html #
Received on Mon Sep 29 1997 - 07:53:49 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:12 MST