Proxmox 1.6 and high cpu usage

V

vitor costa

Renowned Member
Oct 28, 2009
142
2
83
I have a cluster wth drbd and 2 AMD Atholon 4450e, using version 1.5+ 2.6.18. all is ok

A 3 day upgrade to 1.6 and still using kernel 2.6.18 - no problem, all still ok

Today a installed proxmox-pve-2.6.32 and now the cpu usage of vms sky rocket.

The vms used to test is a CentOs 5,5 32 bits and Pfsense firewall (freebsd)

Both processing anything use 1 or 2 % with 2.6.18 and 35% with new 2.6.32-3

Any Idea ?
 
Last edited:
try to go back to the 2.6.18 kernel and see the results. If the issue is reflected to the kernel wait for a new version or install 2.6.32 for KVM Only (do not use it if you have OpenVZ machines).
 
1 - Install proxmox 1.6 in a machine

2 - Create a Vm with full virtualization using 2 or more cpus

3 - Install a operational system (I test with Xp with same results

4 - Alternate the proxmox kernels 2.6.32 and 2.6.18 and view results

I Using webmin + historic system statistics to track cpu usage. Its a nice tool to view cpu for a long time.

I dont try this in intel hardware (i have many servers in production, but only amd to test)
 
Here a CPU grafic showing the cluster node, with 2.6.32 and 2.6.18.

Are 2 VMs runing: Windows XP and PfSense (freeBSD)
cpu18x32-pve1.6.png
 
I update another host to 1.6 wit Intel CPU (Xeon(R) CPU X3430 @ 2.40GHz - Quad-core).

This host have a Windows XP guest + Pfsense Guest (Freebsd).
cpu-0.23_hour.png


The graphic show host cpu usage , first with 2.6.18 (PVE 1.6) and after reboot 2.6.32-3 (PVE 1.6).

The diference:2x more CPU usage with 2.6.32-2 versus when 2.6.18. Apear Intel have a better result... the previous host using AMD have a 5x ratio 2.6.32-3 x 2.6.18 (see previous post). But 2x is a lot more CPU to same use.... I believe previous 2.6.32-2 perform beter...
 
Last edited:
You say in guests ?
The last system :
pfsense has 1 cpu
window xp - 2 cpu

the host have 4 cpu.

I can try this night shutdown the xp vm, so we can see only a uniprocessor pfsense runing with 2.6.18 and 2.6.32


You can reproduce this diference in your envionment ?
 
vitor costa said:
You say in guests ?
The last system :
pfsense has 1 cpu
window xp - 2 cpu

the host have 4 cpu.

I can try this night shutdown the xp vm, so we can see only a uniprocessor pfsense runing with 2.6.18 and 2.6.32


You can reproduce this diference in your envionment ?
Click to expand...

not really. if i run a pfsense and a 2 socket winxp (both without load) I just have a load of 0.01 on my 2.6.32 system (intel quad core)
 
Tom,
Its very strange....
both systems i tested are 1.5 with upgrade to 1.6. Your system is a 1.6 installed from bare metal ? This is no make sense... but....

Whats your processor number (model) ?
 
I installed also the webmin packages and the historical view and I can see a slightly higher load (but not a high load!).

The cpu accounting changed in the 2.6.32 kernel, so its hard to compare results with older kernels.

my cpu: Intel(R) Xeon(R) CPU X3220 @ 2.40GHz
 
You can post your graphic here ?

I will try mount a Proxmox Server using Iso 1.6 an try replicate your results...

you can put your pfsense image to download. So i can try puting this image in my servers and compare with your results... maybe something inside the vm...

If you want i can put my pfsense image to download ....

Thanks for all your attention in this matter.
 
Hey Tom,

I make more tests. Install a fresh 1.5 iso in Athlon 4450e, and fresh install pfsense 1.23 (uniprocessor) and fresh Windows( 1 and 2 cores).Then i upgrade to proxmox 1.6 .. i have take grafics all the road.. and same patern emerge: 2.6.32 use much more cpu with que same guest usage.

In 2.6.18 Windows use more CPU x pfsense
In 2.6.32 Pfsense use more CPU x Windows

Both Guest use more in 2.6.32 x 2.6.18

In pve 1.5 the proporcion is + 70% , in PVE 1.6 is +- 100%( 150% in pfsense case).

Here you can see when i return from 2.6.32-4 to 2.6.18-3 (runing a Windows Xp 2 cores)

cpu-2.6.32-4x2.6.18-3.png
 
You must log in or register to reply here.
Top Bottom