Re: Pid: 8345, comm: rsync Not tainted 2.6.32.22intel #1

[Stefan Priebe - Profihost AG <s.priebe@profihost.ag>]

Hi,

thanks for your suggestions.

 > My other advices : generally don't use cfq scheduler with a RAID
 > controller, it will defeat the whole purpose of RAID cache and command
 > reordering abilities. Use noop, generally, and deepen the queue :
 >
 > echo "noop">  /sys/block/sdc/queue/scheduler
 > echo 512>  /sys/block/sdc/queue/nr_requests

Is there a way to make this static to this disk?

 > Don't hesitate to enlarge tremendously the read-ahead cache, too. I
 > generally use about 512 to 1024 sectors  per drive as a rule of the
 > thumb, so a 4 drives array will use 2048 to 4096 :
 >
 > blockdev --setra 4096 /dev/sdc
Is there also a way to make this static?

 > You should see a 100% write/read speed improvement with these
 > parameters.
That would be great.


Thanks Stefan

Am 26.10.2010 13:25, schrieb Emmanuel Florac:
> Le Tue, 26 Oct 2010 13:03:07 +0200
> Stefan Priebe - Profihost AG<s.priebe@profihost.ag>  écrivait:
>
>> it is a 9650SE-8LPML with Firmware: FE9X 3.06.00.003.
>
> Augh, this firmware is antique :) The latest is 4.10.xx.xx. Anyway,
> don't use firmware before 3.08.xx.xx, there were some nasty bugs.
>
>> So you mean i should upgrade to 4.x Firmware?
>
> Definitely. It will much improve performances, too. Simply download the
> firmware file, extract it, and flash the controller with tw_cli :
>
> tw_cli /cXX update fw=prom0006.img
>
> (the firmware is always in the prom00xx.img file).
>
>> Do i then have to do a
>> filesystem repair? Or just wait if the error accours again?
>
> No, the filesystem will be fine. However you should start a RAID scrub
> with
>
> tw_cli /cXX/uXX start verify
>
> This will rebuild the parity with the new 4.X format (faster writes)
> and help detect any hardware fault.
>
> My other advices : generally don't use cfq scheduler with a RAID
> controller, it will defeat the whole purpose of RAID cache and command
> reordering abilities. Use noop, generally, and deepen the queue :
>
> echo "noop">  /sys/block/sdc/queue/scheduler
> echo 512>  /sys/block/sdc/queue/nr_requests
>
> Don't hesitate to enlarge tremendously the read-ahead cache, too. I
> generally use about 512 to 1024 sectors  per drive as a rule of the
> thumb, so a 4 drives array will use 2048 to 4096 :
>
> blockdev --setra 4096 /dev/sdc
>
> You should see a 100% write/read speed improvement with these
> parameters.
>

_______________________________________________
xfs mailing list
xfs@oss.sgi.com
http://oss.sgi.com/mailman/listinfo/xfs