Xen users - Feb 2009 - Default disk I/O scheduler in linux guest

Hi, all,

I just found out that xen assigns the NOOP disk scheduler for linux guest
OSes. Dom0 uses cfq scheduler (it is the linux default).
Is there a reason for xen to turn off disk request merging in the quest OS
by selecting a NOOP scheduler?
Is it because the request optimization will be performed in dom0 or VMM ?

Thansk in advance,
Jia.


_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Any ideas?

On Wed, Feb 4, 2009 at 5:24 PM, Jia Rao <rickenrao@gmail.com> wrote:
> Hi, all,
>
> I just found out that xen assigns the NOOP disk scheduler for linux guest
> OSes. Dom0 uses cfq scheduler (it is the linux default).
> Is there a reason for xen to turn off disk request merging in the quest OS
> by selecting a NOOP scheduler?
> Is it because the request optimization will be performed in dom0 or VMM ?
>
> Thansk in advance,
> Jia.
>

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

On Wed, Feb 4, 2009 at 5:24 PM, Jia Rao <rickenrao@gmail.com>
wrote:
> Hi, all,
>
> I just found out that xen assigns the NOOP disk scheduler for linux guest
> OSes. Dom0 uses cfq scheduler (it is the linux default).
> Is there a reason for xen to turn off disk request merging in the quest OS
> by selecting a NOOP scheduler?
> Is it because the request optimization will be performed in dom0 or VMM ?
it''s an appropriate default.

everything that virtualizes the IO benefits from using NOOP scheduler.
 the point is that any (re)ordering done by the guest would be useless
when the underlying layers (Dom0 in this case, a SAN block device in
others) mangle the IO requests from several guests.  not only you save
CPU cycles by not trying to be clever on the DomU, but also pushing
the requests as early as possible to the lower layer, the best
optimisations can be done at that layer.


-- 
Javier

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

Javier Guerra wrote:
> On Wed, Feb 4, 2009 at 5:24 PM, Jia Rao <rickenrao@gmail.com> wrote:
>   
>> Hi, all,
>>
>> I just found out that xen assigns the NOOP disk scheduler for linux
guest
>> OSes. Dom0 uses cfq scheduler (it is the linux default).
>> Is there a reason for xen to turn off disk request merging in the quest
OS
>> by selecting a NOOP scheduler?
>> Is it because the request optimization will be performed in dom0 or VMM
?
>>     
>
> it''s an appropriate default.
>
> everything that virtualizes the IO benefits from using NOOP scheduler.
>  the point is that any (re)ordering done by the guest would be useless
> when the underlying layers (Dom0 in this case, a SAN block device in
> others) mangle the IO requests from several guests.  not only you save
> CPU cycles by not trying to be clever on the DomU, but also pushing
> the requests as early as possible to the lower layer, the best
> optimisations can be done at that layer.
>   
If the underlying hardware is RAID5 done in the hardware level, using
cfq instead of noop in the guest gives 2x performance while testing with
a simple dd command.

Noop:

# sync ; date ; time dd if=/dev/zero of=/tmp/test bs=1M count=500 ; date
; time sync ; date
Tue Apr 28 00:03:18 IST 2009
500+0 records in
500+0 records out
524288000 bytes (524 MB) copied, 10.2711 seconds, 51.0 MB/s

real 0m10.395s
user 0m0.004s
sys 0m1.288s
Tue Apr 28 00:03:29 IST 2009

real 0m43.910s
user 0m0.000s
sys 0m0.000s
Tue Apr 28 00:04:13 IST 2009

CFQ in the guest:

sync ; date ; time dd if=/dev/zero of=/tmp/test bs=1M count=500 ; date ;
time sync ; date
Tue Apr 28 00:02:09 IST 2009
500+0 records in
500+0 records out
524288000 bytes (524 MB) copied, 6.44671 seconds, 81.3 MB/s

real 0m6.451s
user 0m0.000s
sys 0m1.204s
Tue Apr 28 00:02:15 IST 2009

real 0m28.451s
user 0m0.000s
sys 0m0.000s
Tue Apr 28 00:02:44 IST 2009

Is noop still the preferred choice to be the default?
>
>   

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users

On Tue, Apr 28, 2009 at 5:50 AM, Sadique Puthen <sputhenp@redhat.com>
wrote:
> If the underlying hardware is RAID5 done in the hardware level, using
> cfq instead of noop in the guest gives 2x performance while testing with
> a simple dd command.
i''ve repeated some quick tests, (i have only an md RAID1, with LVM).
bonnie didn''t report any difference, but dd writes are 35% faster with
''noop'' on the guest than CFQ.

what''s interesting is that the Dom0''s CPU usage is almost half
with
CFQ than NOOP, so my wild guess is that in your case you''re bound by
Dom0 more than DomU


-- 
Javier

_______________________________________________
Xen-users mailing list
Xen-users@lists.xensource.com
http://lists.xensource.com/xen-users