Windows Server 2012 on ProxMox 3.1 - unusual load on the core machine

[Caba]

Hi guys,
I have searched for similar posts in this forum but no success till now.
That's why I decided to register this issue here and to hope that someone could help me.

After installing Server 2012 on ProxMox 3.1 environment i registered an unusual high load on the physical machine.

The config file of VM:
boot: c
bootdisk: virtio0
cores: 4
cpu: host
cpuunits: 1000
memory: 32768
name: Wiho8
net0: e1000=6E:5F:6B:52:FF:7C,bridge=vmbr0
net1: virtio=4E:00:BA:E9:16:2D,bridge=vmbr0
ostype: win8
scsihw: virtio-scsi-pci
sockets: 2
virtio0: proxmox-images:100/vm-100-disk-1.raw,format=raw,size=950G

However, this is the current config, I have made a lot of experiments - 20 cores (2x10), 16 cores (4x4), fixed RAM - 32, 48, 60 GB, Dynamic RAM with Balloon Driver and Balloon service (from 24 to 48 GB), I disabled also the option
"Use tablet for pointer", disabled Time syncronization in windows... No success.
I use 2 network cards just for testing - now e100 is disabled, the virtio one is enabled. I use the latest VirtIO drivers from RedHat.

This VM is intended to be the only one VM on a core machine with 24 cores and 64 GB of RAM, RAID 10 with SAS disks.

Some more info:
1. The Process Hacker application shows a lot of CPU time spent on Interrupts and DPCs (Deferred Procedure Calls). More CPUs I use, more CPU power is wasted on that.
2. The first row from htop output:

PID USER PRI NI VIRT RES SHR S CPU% MEM% TIME+ Command
36159 root 20 0 32.6G 31.4G 4228 S 238. 49.8 1h21:30 /usr/bin/kvm -id 100 -chardev socket,id=qmp,path=/var/run/qemu-serv
36191 root 20 0 32.6G 31.4G 4228 S 35.0 49.8 9:41.62 /usr/bin/kvm -id 100 -chardev socket,id=qmp,path=/var/run/qemu-serv
36190 root 20 0 32.6G 31.4G 4228 S 33.0 49.8 9:53.84 /usr/bin/kvm -id 100 -chardev socket,id=qmp,path=/var/run/qemu-serv
36186 root 20 0 32.6G 31.4G 4228 R 30.0 49.8 9:50.24 /usr/bin/kvm -id 100 -chardev socket,id=qmp,path=/var/run/qemu-serv
36189 root 20 0 32.6G 31.4G 4228 S 29.0 49.8 9:53.47 /usr/bin/kvm -id 100 -chardev socket,id=qmp,path=/var/run/qemu-serv
36185 root 20 0 32.6G 31.4G 4228 S 29.0 49.8 9:36.83 /usr/bin/kvm -id 100 -chardev socket,id=qmp,path=/var/run/qemu-serv
36192 root 20 0 32.6G 31.4G 4228 S 26.0 49.8 10:02.87 /usr/bin/kvm -id 100 -chardev socket,id=qmp,path=/var/run/qemu-serv
36187 root 20 0 32.6G 31.4G 4228 S 24.0 49.8 9:34.55 /usr/bin/kvm -id 100 -chardev socket,id=qmp,path=/var/run/qemu-serv
36188 root 20 0 32.6G 31.4G 4228 S 22.0 49.8 9:40.86 /usr/bin/kvm -id 100 -chardev socket,id=qmp,path=/var/run/qemu-serv



this first row is for the whole VM, all the others are related to each CPU-core, dedicated to the VM. From time to time this CPU-core usage increases till 1500-1800% while all the others are normally used - under 100% each of them.

I think this has something to do with the interrupts and latency on the VM.

Has anyone faced such a problem?

I will highly appreciate any opinion or advice here.

Thanks in advance!

 

[term]

Is this VM running SQL server by any chance?

Does performance improve if you cut back to one socket?

 

[Caba]

Yes, it runs MS SQL SERVER 2012. However I tested the server with stopped all sql services. No success.

 

[term]

I'm not sure it will make a difference, but I would not use the latest virtio drivers. I'm using virtio-win-0.1-59.

 

[term]

I'm running a windows server 2012 vm, 8GB Ram assigned, 8 cores, 1 socket. Process hacker states that interrupts are using .23 % of cpu. The kvm itself is using 5.7% cpu on the host. I have sql server running but its not under load at all.

What happens if you run just one core?

 

[Caba]

I switched to 1 socket, 8 cores.
The same behavior, Interrupts and DPCs still consume up to 25% of the CPU and the load of the physical machine is around 2.
Now I will change the VirtIO drivers to an older version. Now I use 0.1-65.

 

[Caba]

Hi again guys,
I think the problem is in one of the drivers or in any incompatibility between Server2012 and ProxMox.
I am attaching a result of a latency checking. Despite of changing the VirtIO drivers from different builds I ever get this result.

Any ideas?

 

[Patschi]

I have four Windows Server 2012 R2 running on my server (with the latest virtio driver) and don't have any high load you experience. Try to change the CPU-type (e.g. host - what I currently use). Does a clean-installation of WS2012 also have such a high latency?

 

[Caba]

Hi, thanks for the answer!
The CPU type now is host. I also tried with the default cpu type, no difference.
The load and the latency were also so high on a clean installation.

 

[Patschi]

Which Windows Server version you have? Try R1 or R2 - maybe it's a Windows related issue? You could try other Windows versions like XP, Windows 7/8. Is the CPU usage in the VM also so high as on the host machine or just terrible slow?

 

[Caba]

It is not R2 and unfortunately I cannot make tests on this host-machine, I simply don't have place for another VM...
Also this is a production Windows server with a lot ow web-sites and it is very hard for me even to do any changes on the drivers etc.

Any other ideas?

 

[Patschi]

Sadly no. Maybe you can get some free ressources for a Windows XP machine - XP does not need so many ressources.
Does the virtual machine itself use so many CPU or HDD ressources or does just the host use so much?