Subject: Re: Disk performance hit with Dom0 vs native?
To: Manuel Bouyer <bouyer@antioche.eu.org>
From: Stephen Borrill <netbsd@precedence.co.uk>
List: port-xen
Date: 11/17/2005 17:02:38
On Thu, 17 Nov 2005, Manuel Bouyer wrote:
> On Thu, Nov 17, 2005 at 03:56:36PM +0000, Stephen Borrill wrote:
>> This might be just noise actually. I've noticed the performance plummets
>> at around 20G through the disk (80G) to around 11MB/s from much highers
>> speeds at either end. I think it could well be down to a duff disk (but
>> there are no errors in dmesg).
>
> With modern disks, speeds are very different depending on the place of the
> data on the disk:

That's interesting, I never knew that.

> pop# dd if=/dev/rwd0d of=/dev/null bs=64k count=1000
> 1000+0 records in
> 1000+0 records out
> 65536000 bytes transferred in 2.447 secs (26782182 bytes/sec)
> pop# dd if=/dev/rwd0d of=/dev/null bs=64k count=1000 skip=300000
> 1000+0 records in
> 1000+0 records out
> 65536000 bytes transferred in 2.607 secs (25138473 bytes/sec)
> pop# dd if=/dev/rwd0d of=/dev/null bs=64k count=1000 skip=500000
> 1000+0 records in
> 1000+0 records out
> 65536000 bytes transferred in 2.999 secs (21852617 bytes/sec)
> pop# dd if=/dev/rwd0d of=/dev/null bs=64k count=1000 skip=600000
> 1000+0 records in
> 1000+0 records out
> 65536000 bytes transferred in 4.839 secs (13543294 bytes/sec)
>
> This is a 40GB drive, so skip=600000 is near the end.
>
> So it depends on how you did your tests (did you use rwd0d in both cases ?).

I didn't actually do them as a speed test, I was just blanking the drive 
(no count specified) and checking it was OK. Therefore, it just depended 
on how long I waited before doing a kill -INFO!

>> On an on-thread note, installed xen with 3.0_BETA for the first time this
>> morning and it works brilliantly! Thanks to all concerned!
>>
>> Can someone clarify the state of Xen 2.0 vs NetBSD 2.x and 3.0? I thought
>> that NetBSD 2.x supported Xen 1.2 and support for 2.0 was added in 3.0 and
>> -current (with support for 1.2 being dropped).
>
> Yes, that's it.

Good!

>> However, there's xentools20-2.0.7 available as a binary package for 2.0,
>> so I'm confused. Perhaps some explanation should be added to the
>> (excellent) HOW-TO?
>
> Ha, that's a mistake. The package's Makefile doesn't check the NetBSD
> version, and xentools20 doens't need anything but python to
> build, so it can be build on any NetBSD version supporting python.
> But it won't be usefull on anything not i386, and anything < 3.0_BETA.

It does specify i386 (and specifies a version for Linux), but I suppose 
ONLY_FOR_PLATFORM just needs a quick fix to stop people like me getting 
confused. :-)

Thanks,

-- 
Stephen