On Tuesday 23 September 97, at 20 h 40, the keyboard of Paul A Vixie
<paul@vix.com> wrote:
> wierd. i guess i just don't know how to market things. here i've been
> selling transparent caching to real customers since february and nobody
> knows about it, whereas cisco hasn't sold it to anybody but everybody
> knows about it. we're in beta testing for V2.0 of our product, a total
There is at least one interesting thing in the Cisco setup, which is *possibly* translatable into Squid.
They can run a farm of caches, with automatic specialization. Such a farm is difficult because you either have to duplicate the cache on all machines, wasting disk space, or you have to specialize them "by hand" like NLANR does.
On the other hand, the Cisco router computes a hash from the HTTP server name and uses this hash to direct to one cache of the farm. Thus, requests to one given HTTP server will always end to the same cache. No duplication! The router just has to know the IP addresses of the members of the farm.
Translated to Squid, this would be: a farm of unmodified Squid caches and one proxy with a (yet to write) code which would receive clients connections, compute the hash, and proxy them to the right cache. Such a code would have to keep TCP connections, forward UDP requests, add the proper header (Via:?) so we know the real client, compute the hash and that's all.
Does it sound like a good idea?
Received on Wed Sep 24 1997 - 02:06:00 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:37:08 MST