Re: Re : Re: about peerPingTimeOut()

atit_ldce wrote:

> So i think peerPingTimeout() is called for each ICP Time out event.....

Yes.

> So call for peerGetSomeNeighborReplies() is unneccessary only in
> case where we got ICP time out from all peers

Yes.

> in all other case call of peerGetSomeNeighborReplies() will use
> the ICP Reply message received from some peer(s) to make peer
> selection decision..

yes.

> and befoe doing ICP Ping we also check for neighbors which are up
> and if some found up there is a rare chance for them not to
> respond within time limit..

Yes, but it is not that rare with timeouts or lost ICP replies. Not only
is UDP unreliable, Squid also aggressively tunes the timeout to not have
to wait for the slowest peers all the time when you have a mix of fast
and slow peers. The default timeout is calculated as twice the average
parent roundtrip time (or sibling roundtrip time if there is no alive
parents). See neighborsUdpPing().

> So coclusion we have always high probabilty for ICP Reply...

Yes, but also a relatively high probabity of ICP timeout.

> Am i making sense?

Yes.

> I would like to send u following diagram for review....
> 1)Peer selection process
> 2)ICP Communication Handler
> 3)STD for peer selection Process
> 4)STD for Request forwarding..
>
> will u review others ones?

I am happy to review any diagrams or documentation of the peer selection
process you make, assuming you are making this documentation available
to the public.

Regards
Henrik
Received on Tue Apr 01 2003 - 01:07:05 MST