P.V.Anthony wrote: > Hi, > > Got the following error and it is restarting the server. > Not sure how to solve it. Guessing that the ram might be faulty. > > Mar 21 02:27:53 cluster1 kernel: __alloc_pages: 0-order allocation > failed (gfp=0x1d2/0) > Mar 21 02:27:56 cluster1 kernel: __alloc_pages: 0-order allocation > failed (gfp=0x1d2/0) > Mar 21 02:27:56 cluster1 kernel: VM: killing process watchdog
It won't be your ram. This is normal for a 2.4 kernel on a machine that is under memory pressure - possibly a driver or the network stack is trying to allocate memory but interrupts are disabled, so it needs to kill something because it can't reclaim any buffers (as they most likely all dirty) as paging out can't be done while interrupts are disabled. This is probably happening when the machine is busy with disk writing and network IO? This can even happen if a machine apparently has plenty of memory, especially with 2.4 kernels. i.e. even when there is no swap activity. It is to do with burst requirements for free memory and having lots of dirty buffers. > Need advice on how to solve this. > > Fedora Core 1 > kernel 2.4.27 > 32bit We have seen quite a bit of this with machines running 2.4 kernels. You could bump up /proc/sys/vm values for free pages and tune the dirty write out ratio (the minimum watermark before dirty buffers are written out). I can't remember what the exact proc files are for 2.4 - googling will help or looking at /usr/src/linux/Documentation/filesystems/proc.txt Upgrading it to a 2.6 kernel would most likely make it go away. It has many improvements in its VM. _______________________________________________ Slugnet mailing list [email protected] http://www.lugs.org.sg/mailman/listinfo/slugnet
