[ovirt-users] Bad performance with Windows 2012 guests
dfediuck at redhat.com
Sat May 2 08:49:57 UTC 2015
On Apr 30, 2015 14:03, Martijn Grendelman <martijn.grendelman at isaac.nl> wrote:
> Ever since our first Windows Server 2012 deployment on oVirt (3.4 back
> then, now 3.5.1), I have noticed that working on these VMs via RDP or on
> the console via VNC is noticeably slower than on Windows 2008 guests on
> the same oVirt environment.
> Basic things like starting an application (even the Server Manager that
> get started automatically on login) take a very long time, sometimes
> minutes. Everything is just... slow.
> We have recently deployed Microsoft Exchange on a Windows Server 2012
> guest on RHEV, and it doesn't perform well at all.
> I haven't been able to find the cause for this slowness; CPU usage is
> not excessive and it doesn't seem I/O related. Moreover, other types of
> guests (Linux and even Windows 2008) do not have this problem.
> We have 3 different environments:
> - oVirt 3.5.1, on old Dell servers with Penryn Family CPUs with fairly
> slow storage on replicated GlusterFS, running CentOS 6.6
> - oVirt 3.5.1, on modern 6-core SandyBridge servers with local storage
> via NFS, running CentOS 7.0)
> - RHEV 3.4.4 on modern 10-core SandyBridge servers with an iSCSI SAN
> behind it, running on RHEV Hypervisor 6.5
> All of these -very different- environments expose the same behaviour:
> Linux, Windows 2008 fast (or as fast as can be expected given the
> hardware), Windows 2012 painfully slow.
> All Windows 2012 servers use VirtIO disk and network. I think all
> drivers are from the virtio-win-0.1-74 ISO.
> Does anyone share this experience?
> Any idea why this could happen and how it can be fixed?
> Any other information I should share to get a better idea?
> Btw, for the guests on the RHEV environment, we have a case with RedHat
> support, but that doesn't seem to lead to a quick solution, hence I'm
> writing here, too.
> Thanks for any help.
> Martijn Grendelman
Can you please provide the QEMU command line, together with kvm and qemu version?
This information will be helpful for reproducing the problem. However, if the problem is not reproducible on a local setup, we will probably need to ask collecting some performance information with xperf tool.
More information about the Users