tir, 15.02.2005 kl. 07.34 skrev Wes Shull: > Ok, following up on this, I dropped in a new video card this weekend > (generic Radeon 7000), and my system crashes have gone away. So the > video card was indeed the culprit, must have been locking the PCI bus > or something. > > Now I can leave mplayer looping overnight or run folding@home without > any problems. However... > > With the latest rawhide kernels, I still can't run azureus for very > long. Not a crash, but the OOM killer seems to take it (and other > things) out abnormally early. The thing is, it happens in situations > where, near as I can tell, I'm not that low on memory. I do have only > 256 MB RAM, and the kernel slab debug is using a lot of memory as I > documented in an earlier post to fedora-devel-list, but it seems like > the swap (and I've got 512 MB of it) has hardly been touched when this > happens. I am running with vm.swappiness=0, but that's supposed to > just keep it from swapping until it *has* to, not stop swapping > altogether (and it isn't). > > Actually, I just had the OOM killer take out firefox while I was > bugzilla'ing something else :( Here's what it has to say for itself. > This is under kernel-2.6.10-1.1141_FC4(.i686): > > http://kuoi.asui.uidaho.edu/~wes/oom-sucks.txt Yeah, the OOM is "a bit" rash and stupid. I had a pc which i sent a 25 pages pdf (or was it ps) document to the printer - which was controlled by gimp-print. Needless to say, the system ran out of its 128 MB's of RAM in about 2 secounds, trashing the harddrive for about 10 minutes, killing off about *everything* (including gnome dock etc.) - *EXEPT* the wild goose gs (ghostscrip) prosess eating 100 MB's of RAM and 500 MB's of swap. I eventually resorted to the power-cutting method of killing the process... When speaking of videocards and instability - i had a Vodoo PCI card which had the bad habbit of suddenly resetting itself without warning. I cant forget the first time it happened - i was showing some techy Linux for about the first time. I was really, really impressed, and i was just coming around to "stability". Then it happened. The monitor (i have 10 kg's of glass and vacum sitting on my desktop, its a shame the vacum dosn't lift the monitor more...) made a *click*, and went to black with rolling stripes on the screen. I thing about the only open app was evolution... The pc didn't actually *crash* - it just locked up the PCI bus or something. At least it seemed like it was probing for HW all the time - the floppy/cdrom blinked (just as it does when inserting a usb mass storage device, and hotplug loads the mass storage driver) constantly - and the switch told a tale of loosing the network connection every 10 secounds... Anyway, i was able to ssh in, and restart the machine. this was during fc1, i.e. the Linux 2.4 days. Kyrre