hi Henrik: I've tried your suggestion, but the problem still exists. this squid use about 8Mbps bandwidth, some of the statistics: when squid started for an hour: client_http.requests = 26.563860/sec server.all.requests = 14.208499/sec select_loops = 268.088343/sec select_fds = 240.931212/sec cpu_time = 23.120000 seconds wall_time = 300.031695 seconds cpu_usage = 7.705853% after running for one day: client_http.requests = 42.807063/sec server.all.requests = 33.702650/sec select_loops = 236.957915/sec select_fds = 365.309147/sec cpu_time = 148.480000 seconds wall_time = 300.184107 seconds cpu_usage = 49.462978% compare to a squid that only cache static html file(this squid has no hyperthread and output about 15Mbps bandwidth): client_http.requests = 202.408967/sec server.all.requests = 44.233108/sec select_loops = 783.316002/sec select_fds = 825.322455/sec cpu_time = 114.030000 seconds wall_time = 300.001531 seconds cpu_usage = 38.009806% any suggestion? On Tue, 29 Mar 2005 12:51:10 +0200 (CEST), Henrik Nordstrom <hno@xxxxxxxxxxxxxxx> wrote: > On Tue, 29 Mar 2005, netcraft wrote: > > > We use squid to accelerate our dynamic pages(resin/jsp), some pages > > are setting to very short expiry time(2min). and some are one day. It works > > good at the beginning, But I meet a strange problem: when squid start, it use > > about 8% of cpu time. the speed is fast. after 2 or 3 days, the cpu usage > > grows up to 49%, and the speed is slow. > > Try the following > > half_closed_clients off > quick_abort_min 0 KB > quick_abort_max 0 KB > > and don't set maximum_object_size_in_memory overly large. > > Regards > Henrik >
