Re: neighbor timeout behavior in 2.2dev4

From: Chris Teakle <>
Date: Tue, 20 Apr 1999 15:21:32 +1000

>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
Received on Mon Apr 19 1999 - 23:09:01 MDT

This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:45:52 MST