On Mon, Mar 04, 2019 at 05:12:40PM +0100, Michal Privoznik wrote:
On 3/4/19 4:19 PM, Igor Mammedov wrote:
Then I'd guess that most VMs end up with default '-numa node,mem' which by design can produce only fake NUMA without ability to manage guest RAM on host side. So such VMs aren't getting performance benefits or worse run with performance regression (due to wrong sched/mm decisions as guest kernel assumes NUMA topology is valid one).
Specifying NUMA distances in libvirt XML makes it generate the modern cmd line.
AFAIK, specifying any guest NUMA -> Host NUMA affinity makes it use the modern cmd line. eg I just modified a plain 8 CPU / 2 GB RAM guest with this: <numatune> <memnode cellid='0' mode='strict' nodeset='0'/> <memnode cellid='1' mode='strict' nodeset='1'/> </numatune> <cpu mode='host-model'> <numa> <cell id='0' cpus='0-3' memory='1024000' unit='KiB'/> <cell id='1' cpus='4-7' memory='1024000' unit='KiB'/> </numa> </cpu> and I can see libvirt decided to use memdev -object memory-backend-ram,id=ram-node0,size=1048576000,host-nodes=0,policy=bind -numa node,nodeid=0,cpus=0-3,memdev=ram-node0 -object memory-backend-ram,id=ram-node1,size=1048576000,host-nodes=1,policy=bind -numa node,nodeid=1,cpus=4-7,memdev=ram-node1 So unless I'm missing something, we aren't suffering from the problem described by Igor above even today. Regards, Daniel -- |: https://berrange.com -o- https://www.flickr.com/photos/dberrange :| |: https://libvirt.org -o- https://fstop138.berrange.com :| |: https://entangle-photo.org -o- https://www.instagram.com/dberrange :|