You might consider enabling Cache Digests (see http://wiki.squid-cache.org/SquidFaq/CacheDigests). If squid is compiled with --enable-cache-digests, you can configure peers to periodically share a hashed summary of cached objects instead of using ICP to check as requests come in. Checking the local RAM digests for several peers is (nearly always) more efficient than sending out ICP requests to the same number of peers and then waiting for responses from all peers. Drawbacks are the overhead to build and transfer digests every X minutes, and also you miss out on hits that would have been successful with ICP, particularly the extremely efficient UDP_HIT_OBJ type :) Kevin