We should check what QEMU hardware are we using, what is supported and what would be the best. For example, we should use QXL instead of std GPU (but that would require using SPICE instead of VNC), we should use correct CPU model (qemu32 isn't really working with Live images, while kvm32 is) and so on.
std
This ticket had assigned some Differential requests: D655
Another question is whether we are want to test 32bit Fedora on 32bit-only CPU (or are we okay to test it on 64bit CPUs, as we are doing already with virt-manager and HW).
Note: I've been going in through the web UI and changing the CPU settings for the last few weeks (my changes will have been wiped whenever anyone runs 'templates', I usually go back in and change it again when I notice). I've been setting it to 1 CPU per VM and CPU model 'host'. Those settings seemed to run somewhat quicker - I think 2 CPUs per guest, along with the CPU time needed for the video encoding, overpowers the box a bit.
I think if we're going to tweak the CPU settings, though, it would make sense to do so on the new deployment once it's up and running, since that will be the long-term one - optimizing for BOS at this point seems a waste of time.
I have nothing against setting CPU model to host - that's what D655 does, but only for 64bit CPUs. It uses kvm32 for 32bit CPUs. There is a question whether we are OK with testing 32bit images on 64bit CPUs or whether we want to do pure 32bit-images-on-32bit-CPUs thing.
host
kvm32
I've read [[ https://fedoraproject.org/wiki/Changes/Server_Drop_i686 | this ]] and it seems to me that we don't have to care about 32bit CPUs anymore - only thing that comes to mind is 32bit upgrade. As of now, I think that I might delete 32bit machine altogether and use 64bit for testing 32bit images.
! In #637#8565, @adamwill wrote: I think 2 CPUs per guest, along with the CPU time needed for the video encoding, overpowers the box a bit.
We could try whether it's better to run fewer clients with 2 CPUs each or more clients with 1 CPU each.
I'm not totally sure of the overall situation wrt. 32-bit support; Server is trying to get rid of it, yes, but there's also Workstation and Cloud and the KDE live spin. As long as some 'blocking' form of Fedora or other is still 32-bit, we should do at least some testing for it, I think.
Yeah, but when we set CPU model to host on both 32bit and 64bit machine, there is no reason in keeping 32bit machine altogether, because 32bit and 64bit are the same. We could keep only 64bit machine and uefi machine and schedule 32bit images on 64bit machine.
We seem to be hitting problems with SPICE errors on f25 which may be related to using qxl + VNC. I'm testing -vga virtio on staging. See https://bugzilla.redhat.com/show_bug.cgi?id=1403343 .
-vga virtio
Log in to comment on this ticket.