This is an archived copy of the Xen.org mailing list, which we have preserved to ensure that existing links to archives are not broken. The live archive, which contains the latest emails, can be found at http://lists.xen.org/
Home Products Support Community News


Re: [Xen-users] streaming server on a virtual machine.

why would fewer i/o's be needed when accessing lvm vs non lvm if both underlying fs was ext3?

i'm not challenging, just asking as this could effect how i look at things, i would ofcourse verify this as well.

you have the awsome advantage of testing intricate setups which is valuable.

- Brian

On Sep 27, 2009, at 7:42 AM, Grant McWilliams wrote:

On Sat, Sep 26, 2009 at 9:48 AM, Brian Krusic <brian@xxxxxxxxxx> wrote:
Hi Grant,

I'm curious about your tests; LVM vs non LVM for the domU.

Was your domU image based (sparse or no sparse) and using an LVM inside or did you set up your domU to have direct disk access?

Was your dom0 LVM based as well?

What non LVM fs were you using, ext2, ext3, XFS?

- Brian

The DomU for the disk-file test was a non-sparse image file set up as a drive with straight partitions inside like this 'tap:aio:/srv/xen/webserver.img,xvdb,w',. The DomU for the LVM test had an LV in the Dom0 set up as a drive in the DomU like this 'phy:/dev/vgsys/lvweb,xvdb,w'

As you probably know that you'd never use a sparse image file for anything performance related (at least until the hard drive was full!). Both were using ext3. All the variables were the same, same OS drive for both (non-sparse file disk image) and the Dom0 was the same. I just replaced the disk file "drive" with the LVM "drive" at the same mountpoint and re-ran the tests. I had to convince SELinux in both cases that the new location was OK for mysql to write to so an SELinux policy was created that allowed this. I'd always had apache running out of an LV but mysql was writing to the Disk File. After I ran these tests I'm moved mysql over to the LV as well.

I would say though that even though the performance between native and real is virtually nothing you still need to pay attention to the CPU utilization differences. Running the db in the VM is still consuming more CPU than if it were on a bare machine.

It's also worth noting that previously I was working on contract on a full suite testing of the various VT technologies and during those tests I documented disk speed tests between Disk files and LVM and to my surprise I found very little in the way of speed increases of LVM over Disk files (from parity to 10% max). This was not expected at all. I believet the speed increase with the LVM volume was due to fewer IOs needed. mysqlbench is very tough on IOs.

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 mailing list
<Prev in Thread] Current Thread [Next in Thread>