"A month of sundays ago Steve Lord wrote:"
> > Jul 4 22:05 bzip2 linux-2.4.6-xfs-07052001.patch.bz2 817Kb
>
> ^^^^^^^^^^^^
>
> Try this one, it applies cleanly and builds without problem for
> me, including the new aic driver and xfs as a module.
Ditto. I didn't apply any further aic patch yet.
My first impression is that there is some instability. I'm running
on a dell smp machine, and with 2.4.3 + xfs I was getting about 1
hang a week (I know about them because the machine doesn't always
reboot but hangs). I had about six today :-(.
reboot system boot 2.4.6-SMP-XFS Tue Jul 10 20:16 (04:15)
ptb ttyp2 :0.0 Tue Jul 10 19:56 - crash (00:20)
ptb ttyp0 :0.0 Tue Jul 10 19:53 - crash (00:23)
ptb :0 Tue Jul 10 19:53 - crash (00:23)
reboot system boot 2.4.6-SMP-XFS Tue Jul 10 19:52 (04:39)
reboot system boot 2.4.6-SMP-XFS Tue Jul 10 19:50 (00:00)
ptb ttyp3 :0.0 Tue Jul 10 19:40 - crash (00:10)
ptb ttyp2 :0.0 Tue Jul 10 19:39 - crash (00:11)
ptb ttyp0 :0.0 Tue Jul 10 19:37 - crash (00:13)
ptb :0 Tue Jul 10 19:37 - crash (00:13)
reboot system boot 2.4.6-SMP-XFS Tue Jul 10 19:36 (00:14)
reboot system boot 2.4.6-SMP-XFS Tue Jul 10 19:03 (00:47)
reboot system boot 2.4.3-SMP-XFS Tue Jul 10 14:57 (04:03)
I don't have a test yet. I get the impression that it hangs under load.
I cut the speed of the 4 scsi lvd drives to 20MHz. I was runing raid5
over the scsi, with lvm managing the raid resource, and inside the lvm
I had a pair of virtual partitions forming raid0 mirrors with nbd
network devices. The xfs is on these.
I'm pretty sure that syncing the higher level raid while the lower
level raid was still syncing hung the machine. Ditto doing a mkraid
on the higher level. Ditto doing it by hand with dd. I was able
to dd to and from each subdevice separately with no problem. I was
able to fsck.xfs the nbd device, and the raid device over it,
but it hung later while doing nothing in particular.
I haven't been able to test the raw 2.4.6 kernel. My impression is that
this is a raid5 problem, but maybe the xfs patch has something to do
with it? I recall accounting changes to make everything be computed in
sectors coming in from xfs. Is all that still OK.
And it was noticable that the scsi resync speed dropped by half over
2.4.3. That usually implies a change from counting in KB to counting in
512B.
I'll try and do more testing tomorrow. Just a first impression ...
Peter
|