Re: neighbor timeout behavior in 2.2dev4

>I have just installed 2.2dev4 on a machine that is a child of a parent
>for the .com domain.
>
>With previous versions I noticed that the child was using the parent much
>more but now instead of using the parent to retrieve a copy of a request
>it is timing out and getting the request itself.
>
>These machines are on the same network on 100mb lines, and I realize
>that squid dynamically figures out timeouts for parents and siblings by
>round-trip-times, but is this algorithm being too stringent because I
>have a fast network connection?
>
>I have the child using 3 other squids as siblings that are not on the same
>network (2 nlanrs and 1 close by).
>
>Would setting the icp_query_timeout to something besides 0 (auto) help?

I had a similar problem in earlier 2.X versions of Squid with proxies
talking to parents in the same machine room. When I upgraded from
1.NOVM.22 to 2.0.RELEASE, the proportion of direct fetches jumped from
around 30% to 60%. Alot of misses were being logged as TIMEOUT_DIRECT.
That was when I hadn't set icp_query_timeout, i.e. allowing Squid to
dynamically calculate timeouts.

When I set icp_query_timeout to 250, the level of direct fetches went
back to normal.

--
Chris Teakle
Network Development, ITS
The University of Qld