Currently, the guest can not know the NUMA info of the vcpu, which will result in performance drawback. For example: Supposing vcpu-a on nodeA, vcpu-b on nodeB, when load balance, the tasks' pull and push between these vcpus will cost more. But unfortunately, currently, the guest is just blind to this.
So here is the idea to solve it. Export host numa info through guest's sched domain to its scheduler. So the guest's lb will consider the cost. These patches include: For guest: 0001-sched-add-virt-sched-domain-for-the-guest.patch 0002-sched-add-virt-domain-device-s-driver.patch For host: 0001-kvm-collect-vcpus-numa-info-for-guest-s-scheduler.patch 0001-Qemu-add-virt-sched-domain-device.patch Please give some comments and suggestion. Thanks and regards, pingfan
