Re: Still not sure how this works...

> >It doesn't, but it does know that it is getting timeouts when it requests
> >pages from that link. And I'll rephrase that a little: "when link B is not
> >supplying pages, for whatever reason"
>
> How do you decide "when link B is not supplying pages?" If I make a request
> for foo.com, and it travels across link B, and foo.com is down so I get
> connection timeout, or connection refused, did *link B* not supply the page?

Well, it didn't supply *that* page so I'd want Cache A to fetch it directly.
I don't need to turn the parenting 100% or 100% off - as long as it
backs off on a page by page basis but still works (albeit with a short
timeout?) it's better than what I have at the moment where a link failure
takes out the service entirely.

Anyway, you already use heuristics (as you posted yesterday) to guess
when other things are screwed up (you mentioned some miss ratio >= 1
as being a trigger, yes?)

> One thing to do, which would be pretty easy, is to configure Squid to
> ICMP ping the far side of link B for every request. This is basically
> what ICP does. It "ICP pings" the parent cache before every request.

That's a good tip. I'll try it. That may be all I need to make this
work in fact.

G
Received on Thu Apr 10 1997 - 22:18:57 MDT