Re: cache checking algorithm question

Graham Toal writes:
> I think I know the answer to this but I'd like it confirmed please...
>
> I've recently allowed access to my cache to another ISP connected to
> the same Sprint router as me. When he sends a UDP request to my cache,
> if I have the page I'll tell him.
>
> But... I have a second cache that I have a direct ether connection
> to, and my two caches run as proxy-only so they can reduce their
> disk space by keeping only one copy on one server of any page that
> either cache receives (one from a sprintlink line, the other from
> an MCI line).
>
> When the other ISP queries me, do I query my local sibling as well or
> does the request stop with me? If the latter (which I suspect is
The request stops at you.

> the case) then the other ISP isn't getting the full benefit of
> my cache. *but* he cannot add my MCI-hosted cache to his siblings
> because the route between him and that cache is very long as opposed
> to the one-hop over the Sprintlink router.

> Is there a way I can improve this situation?
I spoke to Duane about this a while ago... We have a similar situation
where we have 3 caches, all sitting on the same piece of backbone.

Some of our customers are only allowed to use squid in sibling mode
(note parenting against us) and they are only allowed "cache_hits".
Thus we have had to say "add the following 3 lines to your cache config"
and it then sends 3 packets to the caches here... painful, especially if we
add another cache machine, and if their lines are loaded then it slows things
down...

To your problem I can think of a specific solution:

Add routes to his network's routers for your mci cache... this way if they
send packets to/from it they will travel via your network...
munge munge kludge kludge

Duane also said that I should think about using the "cache-redirect" option
in ICP, but it seems like this doesn't exist ;)

Oskar
Received on Wed Jun 11 1997 - 01:41:32 MDT