Re: Is squid unable to handle the load?

From: John A. Lauro <>
Date: Wed, 27 May 1998 15:04:18 EDT

> > If all squid does was read that might be true,
> Ok, squid is doing a read if we have a cache hit, right?
> Squid can only do 100 reads pr. second right?
> With a cache hit rate at 40% this gives a max rate at 140 reqests/s,
> right?

Not all cache hit's cause a read. If you have enough ram for either
the requests to be in squid's memory cache (assuming you aren't
using the stripped version), or the O/S disk cache, then a physical
read isn't needed... Also, some of the requests are just pass
through (cgi, etc...) or a miss, neither of which should require a
read from disk.

If you really wanted high requests/sec, you could probably use 2
boxes, with one (the parent) being as you would normally configure
squid for high speed with lots of disks.

The second, put in lots of ram (at least 1GB, of course 8GB would be
better), and use an OS like Solaris where /tmp is used like a ram
disk+swap... Put your cache directory on /tmp. Most things that
cann't be cached (such as cgi) would bypass the parent, and 600+mb of
RAM cache will give you a decent hit rate. Point all the clients to
this second box, and it will have little disk activity (next to none
after startup if you put the logs on /tmp too)! If this system is
rebooted, the cache will still hit from it's parent...

While some pages may be slowed down by the parent that has to still
do disk I/O, things that cann't be cached would continue going
through the child direct to the site and the speed (or lack) of the
disks drives would not slow it down. This would drop the load of the
parent too...

John Lauro email:
University of Michigan - Flint
Information Technology Services
303 E. Kearsley St. phone: (810) 762-3123
Flint, MI 48502 fax: (810) 766-6805
Received on Wed May 27 1998 - 12:14:09 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:40:21 MST