A change in the disk scheduling code in the 1.3.9x kernels has had an
unfortunate side effect in the swapping code. The effect is that the
kswap daemon will not always wake up the disk IO request queue. A
typical symptom would be an occasional apparent freeze of up to 5
seconds under heavy swapping load.
Please apply the patch below to 2.0.0 to address this.
Cheers,
Stephen.
--
Stephen Tweedie <sct@dcs.ed.ac.uk>
Department of Computer Science, Edinburgh University, Scotland.
----------------------------------------------------------------
--- linux/mm/vmscan.c~ Wed Jun 19 18:38:52 1996
+++ linux/mm/vmscan.c Wed Jun 19 18:40:30 1996
@@ -404,6 +404,7 @@
while (1) {
kswapd_awake = 0;
current->signal = 0;
+ run_task_queue(&tq_disk);
interruptible_sleep_on(&kswapd_wait);
kswapd_awake = 1;
swapstats.wakeups++;