Messages in this thread | ![/](/images/icornerl.gif) | | Date | Fri, 14 Feb 2003 01:58:02 -0800 | From | Andrew Morton <> | Subject | Re: 2.5.60-mm2 |
| |
Dave Jones <davej@codemonkey.org.uk> wrote: > > On Fri, Feb 14, 2003 at 01:31:44AM -0800, Andrew Morton wrote: > > > . Considerable poking at the NFS MAP_SHARED OOM lockup. It is limping > > along now, but writeout bandwidth is poor and it is still struggling. > > Needs work. > > > > . There's a one-liner which removes an O(n^2) search in the NFS writeback > > path. It increases writeout bandwidth by 4x and decreases CPU load from > > 100% to 3%. Needs work. > > I'm puzzled that you've had NFS stable enough to test these.
This was just writing out a single 400 megabyte file with `dd'. I didn't try anything fancier.
> How much testing has this stuff had? Here 2.5.60+bk clients fall over under > moderate NFS load. (And go splat quickly under high load). > > Trying to run things like dbench causes lockups, fsx/fstress made it > reboot, plus the odd 'cheating' errors reported yesterday.
I have not tried pushing NFS with complex access patterns recently.
BTW, there's a little patch in there from Trond which I forgot to mention: it implements sendfile for NFS, so loop-on-NFS works again.
But we have a refcounting bug somewhere:
# mount server:/dir /mnt/point # losetup /dev/loop0 /mnt/point/file # mount /dev/loop0 /mnt/loop0 # umount /mnt/loop0 # losetup -d /dev/loop0 # umount /mnt/point umount: /mnt/point: device is busy
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
| ![\](/images/icornerr.gif) |