KVM Hardware Interrupts

[oddfellow]

On Windows 2003 R2 Enterpise KVM guests (Terminal Server) I observe a high CPU load processing hardware interrupts (see attachment).

Where do those come from?

On similarly loaded machines running directly on physical hardware (ca. 20 Terminal Service users) and other Win2003 guests on ESX 3 VMs I don't observe such high cpu loads processing HIs.

pveversion -v:

pve-manager: 1.4-9 (pve-manager/1.4/4390)
qemu-server: 1.1-8
pve-kernel: 2.6.24-16
pve-qemu-kvm: 0.11.0-2
pve-firmware: 1
vncterm: 0.9-2
vzctl: 3.0.23-1pve3
vzdump: 1.2-5
vzprocps: 2.0.11-1dso2
vzquota: 3.0.11-1

Guest conf:

name: bevia02
ide2: none,media=cdrom
smp: 2
vlan0: e1000=AE:A6:B2:69:2B:CD
bootdisk: ide0
ide0: vm-102-disk.qcow2
ostype: w2k3
memory: 4096
onboot: 0

 

[muni]

Hi,

I experience the same problem, did find a solution to this?

 

[udo]

Hi,
i had with one vm the strange effect, that the windows-guest are very slow and the cpu-usage at 100%. But in the webfrontend from proxmox the cpu-usage are very low.
I don't check for interupts, but after i convert the qcow2-File to a raw-File the vm run like expected.
Perhaps you should try also the raw-format for the vm?

Udo

 

[muni]

thanks for your answer, how do I go about converting?

 

[udo]

Hi,
1. shut down the vm
2. in the example it's vm 126:

cd /var/lib/vz/images/126/
qemu-img convert vm-126-disk-1.qcow2 -O raw vm-126-disk-1.raw
vi /etc/qemu-server/126.conf

change the line with ide0 to

ide0: local:126/vm-126-disk-1.raw

If all ok, you can delete the old qcow2-file

Udo

 

[muni]

Thanks for your help!!!

With your instructions I Discovered that it was already in a Raw format, any other things you think I should examine?
I have got on the same hardware 2 Identical 2003 R2 installations and only one of them is doing the problems.

 

[udo]

Hi muni,
are the two 2003-servers on the same proxmoxhost, or two hosts with the same hardware?

In my case the problem is only io-related. On a testserver i have a fast raid-controller, but only very very slow disks (old sata - not sata-2). The new disk are ordered but not arrieved yet ;-)
When i produce heavy io - a tar from two big raw-partitions with the result of the same fs, the winclient shows 100% CPU but in the proxmox console are nothing - and the vm freeze and struggle... no go.

See attachment - in this case i delete a 140GB raw-file on /var/lib/vz

So, have a look at the io-performance (and differents between the hosts).

Udo

 

[oddfellow]

Are your two machines equally loaded (users, etc.)?

I guess the hardware interrupts come mostly from network I/O and perhaps mouse/keyboad because of the terminal services. Disk I/O is very low on the machine.

The only strange thing is, our VM was on a physical host before I virtualized it, on the physical machine with the same amount of people working on it, the hardware interrupts did not use so much CPU time (around 5 percent or so, and on the KVM machine the interrupt processing now uses 40-50 percent CPU time).