[ Date Index ] [ Thread Index ] [ <= Previous by date / thread ] [ Next by date / thread => ]
I had a problem with a machine running a 2.4.20 kernel a while ago, it has 2GB of RAM, and a process that deliberatly grabs 1.8GB of this (it's a video server cache, pulls video @ 5mbit from one of 8 servers using jumbo frames, and serves it to upto 100 clients using normal ethernet frames. It stores the video in a 126GB file, and stores the most recent stuff in memory). Trouble was, after a few days, the system started swapping, which lead kscand to start gobbling up cpu cycles, and the whole thing ground to a halt. Swap never seemed to be clearing, even though the total memory used by processes was less than the physical memory. Even after stopping almost every process on the box, swap just didn't clear. After reading arround, I decided to put the newest 2.4 kernel on there, so it now runs 2.4.32. Since then, there's no problems with kscand, and although 76K of swap is used, it doesn't seem to be a major problem :) However top, (and snmp), report an insanely high cpu use in the "system" catagory -- 99% on CPU1, 90% on CPU2 (and 10% for top). The "problem machine" is a vanilla 2.4.32 kernel, however the interupt timer has been increased to 1024Hz when I compiled it (specific need for the software). I assume that means the kernel is using cycles like crazy. A near identical system (only 1GB ram) on the original 2.4.20 kernel (1024Hz interupt), has 3%/8% system load, and kscand and kwapd are the busy processes. It has a load average of 1.00-1.20 though (2CPU box). The load average on the machine is 0.00/0.05/0.02, nothing to be scared about. Why does it have such an apparently high cpu use in system? It doesn't seem to be causing any problems, but it'd be nice to know how to find out why it's so high. -- The Mailing List for the Devon & Cornwall LUG http://mailman.dclug.org.uk/listinfo/list FAQ: http://www.dcglug.org.uk/linux_adm/list-faq.html