Re: Parents, Child, and Siblings?? Can't we all just get along...

From: Stewart Forster <slf@dont-contact.us>
Date: Fri, 14 Aug 1998 10:08:25 +1000

> > Questions. I have one child squid (A) with two parents ( B and C ) and
> > parent B has sibling (D).
> >
> >
> > D ---- B C
> > \ /
> > \ /
> > A
> >
> > When squid A asks for a page from the parents, does squid B check with
> > sibling before giving answer or only if it has to actually fetch the page?
> > And if it does ask sibling D first and D has it, does this count as a hit on
> > Parent B from A?
>
> In your diagram A sends ICP requests to C & D. C & D will reply
> via ICP (without contacting ANYONE else) whether or not they have the object
> in question.

        Of course this bit above should have read

        In your diagram A sends ICP requests to C & B. C & B will reply
via ICP (without contacting ANYONE else) whether or not they have the object
in question.

>
> If B returns ICP_HIT, A sends the HTTP request to B, and B will
> satisfy it. Same again for C.
>
> If B & C have both returned ICP_MISS, then A will choose whichever
> responded first to send the HTTP request to. If C got there first, it will
> retrieve the object from the remote site (assumption) and pass it back to A.
>
> If B responded first, the HTTP request comes to B from A. B now has
> a request for an object it doesn't have, so it asks its peers via ICP, ie
> B does an ICP request to D. After that, you know the rest.... If D hits,
> HTTP request from B to D, if D misses, B satisfies via direct source.
>
> In answer to your question, ICP requests are not passed implicitly
> around a hierarchy, they are strictly point-to-point.
>
> Stew.
>
>
Received on Thu Aug 13 1998 - 17:09:58 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:41:37 MST