On 20.11.2014 09:12, Laszlo Hornyak wrote:

Hi list,

Hey Laszlo, long time no see!

According to the libvirt documentation [1]

"Show maximum number of virtual CPUs for guest domains on this connection"

This may not be a complete definition.

My first guess from the above was that this returns the number of
virtual CPUs that can exist at the same time on the host, either one or
several VMs. In fact it returned 16 in my fedora 20 desktop running with
KVM, while this is not the real limit of vCPUs. I can define a VM with
more than 16 vCPUs (which does not really make sense when I have only
two cores, but it looks cool) and  also I can have well over 16 virtual
machines running on the host. I looked into the implementation and now I
understand that each driver returns different values, but I do not
understand the constants. Can you guys give me some hints how should I
understand the output of the command?

Well, qemu drivers serves both virt types: kvm and bare qemu. While the latter has no real limit (it's just an userland process without hw acceleration), the former involves KVM in kernel and thus has some limitation on maximal number of machines (inherited from KVM limitation). So, unless you specify which virt type you want to use, libvirt blindly returns 16 (arbitrary meaningless value). But if you specify kvm you'll get sensible value:

virsh # maxvcpus
16

virsh # maxvcpus --type kvm
255


Hope this shed more light in.

Michal