Re: [Xen-users] disk access besk practice

["Grant McWilliams" <grantmasterflash@xxxxxxxxx>]

On Tue, Jan 6, 2009 at 12:07 PM, Todd Deshane <deshantm@xxxxxxxxx> wrote:

On Tue, Jan 6, 2009 at 1:56 PM, John Madden <jmadden@xxxxxxxxxxx> wrote:
>> While performance seems fine both interactively and using benchmarks,
>> is there a practical limit to the image size before I should start
>> breaking it up?
>
> I do *everything* through physical allocations -- phy: -- and I'm quite happy
> with it.  tap:aio seems like unnecessary overhead.
>
> John
>

It's amazing how little that overhead is though. I've done extensive testing on native LVM, file: and tap:aio and I think most people would be surprised at how well Xen handles disk files. Dom0 LVM is faster than a DomU in a tap:aio file but not in all benchmarks and not by much when it is. Maximum was 20% speed increase over tap:aio in the extreme case. Half the tests show that the speed difference between the two is within the error range of the test. 

Yes phy: is faster but I'd only use it in those cases where I absolutely had to have the last bit of disk performance. It always feels a bit hackish to have your DomUs running out of a disk file but I wouldn't choose another method based on perfomance. Disk files do have other advantages.

All of my test results will be released in my upcoming Virt-Report. I have a lot of work still to do.


Grant McWilliams

Some people, when confronted with a problem, think "I know, I'll use Windows." 
Now they have two problems.
_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-users