Re: [squid-users] Anyone considered this potential feature?

From: Chris Tilbury <Chris.Tilbury@dont-contact.us>
Date: Fri, 23 Oct 1998 14:42:08 +0100

On Fri, Oct 23, 1998 at 01:04:32PM +0000, Michael Sparks wrote:

> Squid 2 already hashes the URL when it gets the request to check its
> digest. It would be useful if Squid could be patched to send a request to
> a particular sibling based on the result of that hash. ie
>
> First ensure Squid2 is running on all the machines, then for each request:
>
> X = result of Squid URL hash for it's digest.
> Y = X mapped over N machines, selecting 1. (ala hash value mod N)
> IF Y=0 then go direct
> ELSE get page from sibling Y (unless sibling fails, then go direct)

Sounds like a peer version of CARP to me. Is that the general idea?

> This removes the need to transmit digests between the hosts since host X
> is always responsible for getting the same page. Obviously a configuration
> file for the machines in the cluster in service would be required. (With
> perhaps a backup host for each result of the hash in case of failure)

Hmm. Do you forsee transmitting the cache digests as being a problem? From
the specs, it doesn't sound as if they would impose an enormous load, and
you can turn off ICP querying of sibling caches in the config file, so you
don't need all that ICP traffic.

> Firstly, can anyone see any fundamental flaws in this approach, and if
> not, has anyone done any patches along these lines? If not where would
> those who have been developing squid suggest I start in order to create
> the patch myself... (Unless there exists some kind soul out there who is
> willing to write it? :)

I'm no expert, but I think this wouldn't be such a difficult thing to
implement. It does sound very much like CARP and most of the code for that
is already there.

Maybe you could have multiple FEP systems, though, perhaps on a DNS RR
basis. Squid is fairly good at detecting when a parent is down, after all,
so your children (us! :-) shouldn't suffer if one of the FEP systems went
away. Or have you already discounted this?

Cheers,

Chris

-- 
Chris Tilbury, UNIX Systems Administrator, IT Services, University of Warwick
EMAIL: cudch+s@csv.warwick.ac.uk PHONE: +44 1203 523365(V)/+44 1203 523267(F)
                            URL: http://www.warwick.ac.uk/staff/Chris.Tilbury
Received on Fri Oct 23 1998 - 07:57:49 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:42:45 MST