On a qemu-kvm hypervisor with a hyperthreaded CPU, does matching the VM's core/thread CPU topology to the host's CPU core/thread topology then pinning of the thread pairs improve performance?
Basically using
Basically using
sysctl -n kern.sched.topology_spec, I noticed that using host-passthrough caused the FreeBSD guest to see 8 "real" cores while switching to host-model and setting the cpu topology let FreeBSD create thread groups. With the host being a Linux system running a qemu-kvm managed by libvirt, my hunch is that it does improve performance because the guest OS can manage it's vCPU resource allocations in a way that "helps" the host by more accuratly matching real resource allocations.