Re: icp and multiple siblings + parent, selection mechanism

marc.van.selm@nc3a.nato.int writes:

>How are ICP-requests to multiple siblings and a parent handled?
>
>I have a setup up with 2 siblings and a parent. If an URL is requested and
>it can't be served from the local cache an ICP-request is send out and
>eventually the URL is fetched from one of the siblings, the parent or direct.
>
>My questions are:
>
>1) Is an ICP-request send to all the siblings and the parent at the same
>time? So the parent and the sibling are handled in the same way at this time
>(ignoring what will happen if all reply with a miss) Or are the probed 1 by
>one? (just guessing)
>
>I ask this question because I got the impression that the siblings get much
>lesser numbers of ICP-requests than my parent. This suggests that the parent
>is preferred and only if this one gives a MISS the siblings are tried.

Assuming you have no 'cache_host_domain' or 'cache_host_acl' configuration
rules, then all your neighbors (parents and siblings) are treated
equally, and all are queried at the same time.

>2) Assuming we get an ICP_HIT from more than one sibling/parent, who gets to
>deliver the data (TCP)? Is this who replies first, is the sequence in
>squid.conf important or is the parent handled differently than the sibling?
>(see question 1)

The one who replies first is used. The order in squid.conf does not
matter because each time we send a set of ICP queries, we cycle
the first neighbor.

Duane W.
Received on Tue May 20 1997 - 10:07:14 MDT