On Fri, 2004-10-08 at 09:18 +0930, Benjamin Close wrote: > I believe the issue is with the scheduler or something related to it. > The effect seems to occur when a process > hasn't been used for a long time and then is suddenly switched from a > wait state to an active state. > One place I notice it is if I su then perform some function, leave the > process su'd say overnight then the next morning type exit on it. > The 'exit' takes a while to display then once it does, it takes 3-4 > seconds before the previous shell's prompt reappears. > The box is never in swap with it's stats being: > > Mem: 133M Active, 236M Inact, 92M Wired, 15M Cache, 60M Buf, 18M Free > Swap: 512M Total, 512M Free > > However, it has a number of constantly running processes (graphical > memory watches, xfishtanks, graphical load watches, etc) > > Once the process responds it runs fine until the next long delay. It > seems more responsive with BETA7 but no where near the same as before > the scheduler change. > It might be a symptom related to X, will test overnight in console. > > Cheers, > Benjamin > I have to "me too" this. I have only seen the problem with unused gettys and bashes on vtys and in xterm. I typically do have a reasonable amount of stuff paged out, but the delays I see aren't consistent with lots of paging in. The system can be idle with no disk activity and I can still see ~4 sec delays. I have tried to get a ddb trace on a frozen process with no luck so far - all I ever end up with is 'Thread xxx not found'. I initially assumed this might be because part of the process was paged out - I am unsure, though. Sam.Received on Fri Oct 08 2004 - 02:28:33 UTC
This archive was generated by hypermail 2.4.0 : Wed May 19 2021 - 11:38:16 UTC