Re: Squid server performance

From: Jon Kay <jkay@dont-contact.us>
Date: Wed, 21 Jul 1999 04:52:58 +0000

kayleigh@morgoth.celcom.com.my wrote:
>
> Hi,
>
> I'm running squid2stable4 on hpux 10.20 with 180mgz cpu, 128 mb of ram,
> 4 GB of scsi disk.The server surrently support more than 500 users at one
> time. I've have no crash problem since I upgraded it to stable 3 and now stable
> 4.
>
> What is the best spec to run squid for this amount of users? 500-600 at
> one time. It is advisesable? or maybe I should have another box and split
> the users.

The relevant threshold to watch is load. I am not familiar with your hardware
configuration, but most reasonably configured squids can handle up to about 100
requests/sec. At a guess, unless your machine is dramatically slower than most
modern machines, the bottleneck will be your disk (you only mention one drive).

Disk space is a consideration, but you will probably need a much bigger user
population to fill 4G (amount of disk space used generally depends on user
population).

You may be able to reclaim part of the disk for other purposes so long as they
impose light load.

It depends on what the users are doing, but you'll almost certainly be OK loadwise
as well. If your users are not usually logged on, then, well, universities with tens
of thousands of students have been known to get along with two Squids.

If they are usually active, but 'normal users' who do not place particular load
on servers, then the following data becomes relevant (I wish I had kept track of
who posted this gem):

> Analysis of end user traffic to proxy servers at America Online done
> at Virginia Tech shows that an average user requests one URL about
> every 50 seconds, or a request rate of 0.02 URLs/second. (This does
> not mean that a person clicks on a link or types a new URL every 50
> seconds. Instead, each URL requested typically embeds other URLs,
> such as images. The average rate of the individual URLs requested
> either by a person or indirectly as an embedded object is one every 50
> seconds.)

It is most unlikely that your users will behave just like AOL users, but it's a
1req/50sec is a useful order-of-magnitude estimate. If we divide 500 users by
50 seconds/req, we get a load of roughly 10 requests/sec, within Squid capacity
by an order of magnitude. And note that that AOL data filters out users who
are not logged on.

-- 
Jon Kay        pushcache.com        jkay@pushcache.com
http://www.pushcache.com/           (512) 420-9025
Squid consultants: configuration, installation, modification, design
Received on Wed Jul 21 1999 - 03:47:49 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:47:28 MST