We upgraded from squid 3.0STABLE8 to squid 3.1.0.13 for some new features in ICAP code. We push relatively high traffic through the squid box(around 800 requests/second for each box with 16 squid instances running). The box has 8 CPUs and 8G memory. Squid cache is disabled but ICAP is enabled. All these work fine on 3.0STABLE8. After we deployed squid 3.1.0.13 on last Tuesday, the throughput (requests/seconds) becomes spiky. For example, the client requests sometimes dropped from 800 requests/second to 200 requests/seconds and recovered after a few minutes. The requests/second is measured using SNMP cacheProtoClientHttpRequests. Anybody has similar problem? Thanks
