Skip to contentSkip to conversation

dividing host CPU performance

DISCUSSION Břeťa Vomočil 3 participants 4 messages Dec 5, 2008 – Dec 16, 2008 2 messages not in the archive KVM

lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

Hi,
I want to divide host CPU performance between 2 VM in specified ratio. I use Cgroups system and cpu.shares files, but I'm not much successful in doing that.

My scenario:
- 2 VM - Ubuntu server 8.10 with 4 VCPU, QCOW2 disk and about 300MB memory running on host CPU with 4 cores
- in Cgroups:
1st VM - cpuset.cpus = "0-3", cpuset.mems="0", cpu.shares="3072", tasks=<appropriate_VM_PID> 2nd VM - cpuset.cpus = "0-3", cpuset.mems="0", cpu.shares="1024", tasks=<appropriate_VM_PID>

Actual result:
Performance isn't divided 3:1 (in favor of the first VM) but 1:1.

I'm looking for an approach which gives me abilities like the Xen has with its scheduling parameters "cap" and "weight". I would appreciate any help how to solve this issue - I don't insist on Cgroups, but it seems like an capable solution, but for me it doesn't work (it is also possible, that my scenario miss some necessary setting).

Thanks in advance for any constructive advices
Breta

A message that is not in the archive (sent off-list or not yet received): [email protected]

#3 lore raw
Quoted 3 lines

If your hardware supports it you could look into:

http://www.linuxmanpages.com/man8/numactl.8.php

Yes, my HW support it (I have already used it for keeping a VM on specified CPU core), but it doesn't solve the "generic ratio problem". For this case it can be usable - ratio 3:1 on 4-core CPU is simple task with numactl, I agree, but I'm looking for some more generic solution where you can make different ratios, which is not possible to do with numactl (for instance 4:1 on 4 cores).

#5 lore raw
In reply to Břeťa Vomočil

Břeťa Vomočil napsal(a):

Quoted 20 lines

Hi,
I want to divide host CPU performance between 2 VM in specified ratio. I use Cgroups system and cpu.shares files, but I'm not much successful in doing that.

My scenario:
- 2 VM - Ubuntu server 8.10 with 4 VCPU, QCOW2 disk and about 300MB memory running on host CPU with 4 cores
- in Cgroups:
1st VM - cpuset.cpus = "0-3", cpuset.mems="0", cpu.shares="3072", tasks=<appropriate_VM_PID> 2nd VM - cpuset.cpus = "0-3", cpuset.mems="0", cpu.shares="1024", tasks=<appropriate_VM_PID>

Actual result:
Performance isn't divided 3:1 (in favor of the first VM) but 1:1.

I'm looking for an approach which gives me abilities like the Xen has with its scheduling parameters "cap" and "weight". I would appreciate any help how to solve this issue - I don't insist on Cgroups, but it seems like an capable solution, but for me it doesn't work (it is also possible, that my scenario miss some necessary setting).

Thanks in advance for any constructive advices
Breta

Quoted 5 lines

--
To unsubscribe from this list: send the line "unsubscribe kvm" in the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html

Some time passed without any solution similar to the complex cgroups (thanks for all replies)... could you please advise me, which mail conference would be better to ask for that? It is very important for me to solve this issue...

Thanks
b.