On Wed, 7 Dec 2011, Ryan C. England wrote:
> I have included the current value of vm.min_free_kbytes below.
> [root@localhost ~]# cat /proc/sys/vm/min_free_kbytes
> 90112
We typically (at SL 5 though) double the anaconda-configured value.
It is easy enough to change and test dynamically ..
> On Wed, Dec 7, 2011 at 1:58 PM, Ryan C. England <[log in to unmask]>
> wrote:
> Denice,
> First and foremost, thank you for the reply. I have included the I/O
> stack for your convenience.
>
> nfs, xfs, lvm, megaraid_sas
>
> I could get rid of lvm layer, but the others are required. We're
> currently using this as our NAS. When the array was created, ext4 was
> not yet widely adopted so we chose to go with xfs.
>
> What value should I set for vm.min_free_kbytes? In further
> researching the issue, I have discovered that 2.6.35 includes "xfs:
> skip writeback from reclaim context." Should I entertain the idea of
> using a newer version of the kernel?
probably only if you are comfortable doing it, and have some kind of a
test system.
I would join the XFS mailing list, and ask some experts.
http://oss.sgi.com/mailman/listinfo/xfs
At SL6 I think you can enable some debugging by mounting debugfs
and get some stack traces (some one is probably going to ask for them).
This seems to work for me, but your mileage may very:
# ls /sys/kernel/debug/
# mount -t debugfs nodev /sys/kernel/debug
# ls /sys/kernel/debug/
bdi dynamic_debug kprobes sched_features x86
boot_params extfrag kvm tracing xen
dri hid mce usb
# cat /sys/kernel/debug/tracing/stack_trace
Depth Size Location (-1 entries)
----- ---- --------
#
# Stack tracer disabled
......
# echo 1 > /proc/sys/kernel/stack_tracer_enabled
# cat /sys/kernel/debug/tracing/stack_trace
Depth Size Location (42 entries)
----- ---- --------
0) 3200 48 walk_tg_tree+0x6a/0xb0
........
cheers, etc.
--
deatrich @ triumf.ca, Science/Atlas PH: +1 604-222-7665
<*> This moment's fortune cookie:
To love is good, love being difficult.
|