Re: [squid-users] Performance questions....

On Wed, 2002-11-27 at 23:13, Perry, Owain (Gamer.tv) wrote:
> Hello,
>
> I have been asked by the boss to work out how many users we can serve
> from a squid server running as an accelerator. we are using squid 2.5
> stable 1 on Solaris 8 running on an E450, The website is not to big 50M
> max, were running squid from memory so there is no disk I/O to consider.
> I know this is a tricky thing to work out so my questions are more of a
> theoretical nature than actual figures:
>
> 1. How dependant on hardware will the number of requests serviceable in
> a second be? is memory and CPU going to be more of a bottle neck, or
> will the squid software have it's own maximum limit here?

Actually, the poll() system call is most likely the limiting factor.

> 2. Will this be also be dependant on the maximum number of TCP
> connections that the OS can support?

Yes.

> 3. If a TCP connection is in a Time_Wait state does that count as a
> connection to the OS?

Yes.

> 4. I assume that a squid object is each individual item e.g. a single
> gif file, css file or a html page etc.

Yes.

> How many TCP connections to the
> back end server would squid make for an un-cached page? would it be 1
> connection per item on that page? or 1 connection for all the items.

at most one connection per client request. Connections will be reused
where possible.
 
> 5. Has anyone worked out this kind of data and have any results they
> would like to share?

Sorry, no hard figures from here.

Cheers,
Rob