I have been using Squid for several months on a variety of platforms,
and have run into one oddity -- it appears that Squid's (2.5STABLE1)
internal formula for calculating the ICP timeout value often comes up
with an unreasonably low value. There is a squid.conf variable which
adjusts the 'maximum_icp_query_timeout', but no equivalent option to
set a minimum value?
Adding the appropriate code to set the minimum in 'squid.conf' is a
trivial change (this feature is available in SQUID3), is there a patch
for 2.5STABLE1 to add this feature? Should I publish my patch?
Details:
Each of my 'child' caches cannot access the Internet directly, and is
configured to use ICP to contact each of the available 'parents',
with the 'round-robin allow-miss' option on all cache_peer lines. We
noticed a few weeks ago the child caches log a large percentage of their
requests as 'TIMEOUT_ROUNDROBIN_PARENT', even though a network packet
sniffer shows that all the parents would respond to ICP queries within
30-100 msec.
When the 'slow' response a just another MISS this is not a big deal,
when the ignored response would have resulted in a PARENT_HIT, the
excessively small value of the calculated ICP timeout becomes a problem.
Setting a minimum value for the calculated ICP timeout eliminates nearly
all of the timeout log entries -- it doesn't appear to have significantly
improved the hit ratio.
Kevin Kadow
Received on Tue Dec 03 2002 - 19:54:52 MST
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:19:00 MST