Martin Hamilton writes:
>We see the scenario that from time to time the amount of traffic
>flowing through a particular cache server exceeeds what it's able to
>deal with. OK, so we reroute some of the traffic through other
>servers, but in the meantime it would be useful for the overloaded
>cache to be able to either refuse connections or (perhaps) return an
>ICP message telling peering servers not to talk to it for a while.
returning an ICP_MISS_NOFETCH message is easy. The hard part is
figuring out when to start sending those instead of ICP_MISS.
We already do this with the checkFailureRatio() function and
the 'hit_only_mode_until' variable.
Outright refusing TCP connections is a harsher solution
I think. Client caches with the connect-retry code
will just repeat the connection attempt a few times, and
then report an error.
>I'm thinking that this could be based on something like the length of
>time taken to deliver a cached object to a client, and/or the number
>of requests per second being dealt with - configurable via squid.conf.
sure, that would probably work. Would be nice to make it
automatic though.
Duane W.
Received on Tue Jul 29 2003 - 13:15:49 MDT
This archive was generated by hypermail pre-2.1.9 : Tue Dec 09 2003 - 16:11:47 MST