A couple of updates today... We do plan to give the logfile-daemon approach a try. There's some concern that squid outright kills itself if one of the logfile daemons exits. That creates some potential for data loss, but we're not sure how squid could handle it significantly better without major, complex changes. With mem-only caching it restarts pretty much instantly, and since that should never actually happen, it should be OK. We can live with losing a few K of logs in the event of a crash as long as the crashes are rare. :-) It sounds like we'll need to write our own logfile daemon to handle our specific rotation needs. That should enable us to eliminate the daily squid rotation and restart it only for problems or reconfigurations. Thanks again for the suggestion! On the superpages... this worked very well right up until the actual rotation: Thu Nov 26 04:51:03 UTC 2009 vm.pmap.pv_entry_count: 61361 vm.pmap.pde.promotions: 23123 vm.pmap.pde.p_failures: 327946 vm.pmap.pde.demotions: 17848 --- rotate starts --- Thu Nov 26 05:01:03 UTC 2009 vm.pmap.pv_entry_count: 1480641 vm.pmap.pde.promotions: 23159 vm.pmap.pde.p_failures: 343553 vm.pmap.pde.demotions: 20637 --- rotate finished --- Thu Nov 26 05:11:03 UTC 2009 vm.pmap.pv_entry_count: 1908056 vm.pmap.pde.promotions: 23212 vm.pmap.pde.p_failures: 413171 vm.pmap.pde.demotions: 21470 Ever since then, it's gone back to the old way, with lots of constant p_failures and frequent demotions. Note the machine has 10GB free RAM at the moment, so there's no memory pressure. Currently: Thu Nov 26 13:31:04 UTC 2009 vm.pmap.pv_entry_count: 2021342 vm.pmap.pde.promotions: 24950 vm.pmap.pde.p_failures: 941176 vm.pmap.pde.mappings: 1641 vm.pmap.pde.demotions: 21680 Thu Nov 26 13:41:04 UTC 2009 vm.pmap.pv_entry_count: 2021053 vm.pmap.pde.promotions: 24973 vm.pmap.pde.p_failures: 946147 vm.pmap.pde.mappings: 1641 vm.pmap.pde.demotions: 21681 So it seems like at present the superpages feature is great right up til you need it, unless squid does something odd to its memory at rotation time. I will report this to the freebsd-hackers list as a possible problem with superpages, and see if they have any thoughts. Thanks again everyone!
