On Wed, May 18, 2016 at 8:01 PM, Karli Sjöberg <karli.sjoberg@slu.se> wrote:

>> > I tried a vanilla centos-7.1 as well and the same happens. I'm of the same opinion that this is more a guest related issue, it's just I'd like to find out why this only happens with QXL and not with CIRRUS.
>> >
>> > Thanks.
>>
>> Very interesting. Are your hosts all of the same architecture(family)?
>>
>> /K
>
>
> Actually we have a nice mixture of manufacturers. We run 7 hosts, paired 4-2-1 in relation to architecture. All of them have the same resources, though (CPUs + RAM). In this case, I can't test the VM on different hosts because we've separated one of them (one of the "4") on a standalone oVirt datacenter as we're making tests on it, but I remember this has already happened to me in the past (I didn't have the time to debug it at that time, though).

And it doesn't happen with a similar VM in the other datacenter, on the same hardware?

/K

Hello,

I would like to come back to the original post to get some clarifications:

you write
"

I've deployed a pretty basic VM (ubuntu 14.04 server, 4GB RAM, 4 CPUs, 15GB storage).
"

It seems to me not so basic a VM with 4 vcpus for a desktop.... what are your needs for it?

In general a multi cpu VM could have a worse performance than a mono cpu one, depending on its actual workload, because from the host side there is a continuous scheduling of these 4 vcpus among the real cpus/cores/threads.

I worked in the past with a 4 vcpu VM and 24Gb of ram (in vSphere) used as JDEdwards database server with hundreds of concurrent users ...

Also with JBoss VMs configured with 4-6 vcpus .. But for a basic VM seems somehow overestimated...

In my case on a nuc6i5 with 2 cores + ht and 32Gb of ram, configured as hypervisor with self hosted engine VM, I have a CentOS 6 VM configured with 3Gb of ram and 1 cpu and it runs quite smooth from inside a spice session configured with resolution of 1680x1050 and its cpu usage shows 98.5% idle.

Also from the host point of view

   PID USER      PR NI    VIRT    RES    SHR S %CPU %MEM     TIME+ COMMAND
16025 qemu      20   0 8988404 3.070g 10524 S   1.7 9.9 90:16.07 qemu-kvm   <--- engine VM
24367 qemu      20   0 3852288 1.929g 12188 S   0.0 6.2 41:12.79 qemu-kvm   <---- CentOS 6 VM

If I add to the mix a CentOS 7.2 VM configured with 1 vcpu and 3Gb of ram and with a classic gnome desktop session and open it with 1680x1050 resolution too in spice, I can work without problems and with great fluidity. I see inside vm its cpu jumping from 1% to 6% because of gnome-shell.

I see this at host side while in CentOS 7 VM the packagekitd daemon is downloading the updates and so it is comsuming cpu cycles, also due to virtual network activity

PID USER      PR NI    VIRT    RES    SHR S %CPU %MEM     TIME+ COMMAND
16025 qemu      20   0 9004796 3.077g 10524 S   2.0 9.9 90:27.34 qemu-kvm
24367 qemu      20   0 3852288 1.929g 12188 S   0.7 6.2 41:17.35 qemu-kvm
31277 qemu      20   0 4121224 2.116g 12056 S 16.3 6.8   2:13.80 qemu-kvm

Otherwise very few cpu usage, after packagekitd completes its work;

PID USER      PR NI    VIRT    RES    SHR S %CPU %MEM     TIME+ COMMAND
16025 qemu      20   0 9004796 3.077g 10524 S   2.3 9.9 90:34.42 qemu-kvm
24367 qemu      20   0 3852288 1.929g 12188 S   0.7 6.2 41:20.42 qemu-kvm
31277 qemu      20   0 3834364 2.138g 12056 S   1.0 6.9   2:28.80 qemu-kvm

HIH debugging,

Gianluca