Re: [Xen-users] Disk IO tuning

["Fajar A. Nugraha" <list@xxxxxxxxx>]

On Fri, Dec 16, 2011 at 4:27 AM, Andrew Wells <agwells0714@xxxxxxxxx> wrote:
> So with the the pv
>
> My writes are alright, could be better,
>
> but my reads are really slow compared to the hardware system?
> Any hints for tuning the pv disk io?

How did you measure "really slow"? Are you sure it's not the effect of cache?

I suggest you try again using block device directly (without
filesystem). For example, this is on my system (RHEL6 + xen-4.1.1+
kernel 3.1.0 + zfs), testing using "dd_rescue -d -b128k -B128k -m128M
/path/of/block/device/tested /dev/null", testing the same block device
in three scenario:
- on dom0 directly: got about 22 MBps (yes, this is pretty slow. I
only have two disks, mirrored).
- on dom0, but with block device mapped to dom0 with "xm block-attach"
(not sure what the equivalent command is in xen server): about 18 MBps
- on PV domU: about 18 MBps

the "-d" flag to dd_rescue opens the block device using O_DIRECT, to
avoid cache skewing up the result.

-- 
Fajar

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