Message ID | 20200523232035.1029349-1-peterx@redhat.com |
---|---|
Headers | show |
Series | KVM: Dirty ring support (QEMU part) | expand |
On Sat, May 23, 2020 at 07:20:24PM -0400, Peter Xu wrote: > I kept the dirty sync in kvm_set_phys_mem() for kvmslot removals, left a > comment on the known issue about strict dirty sync so we can fix it someday in > the future together with dirty log and dirty ring. Side note: patch 3,5-8 should not be RFC material at all - they either fixes existing issues or clean code up. Please conside to review/merge them first even before the rest of the patches. Thanks,
On Sat, May 23, 2020 at 07:20:24PM -0400, Peter Xu wrote: > I gave it a shot with a 24G guest, 8 vcpus, using 10g NIC as migration > channel. When idle or dirty workload small, I don't observe major > difference on total migration time. When with higher random dirty > workload (800MB/s dirty rate upon 20G memory, worse for kvm dirty > ring). Total migration time is (ping pong migrate for 6 times, in > seconds): > > |-------------------------+---------------| > | dirty ring (4k entries) | dirty logging | > |-------------------------+---------------| > | 70 | 58 | > | 78 | 70 | > | 72 | 48 | > | 74 | 52 | > | 83 | 49 | > | 65 | 54 | > |-------------------------+---------------| > > Summary: > > dirty ring average: 73s > dirty logging average: 55s > > The KVM dirty ring will be slower in above case. The number may show > that the dirty logging is still preferred as a default value because > small/medium VMs are still major cases, and high dirty workload > happens frequently too. And that's what this series did. Two more TODOs that can potentially be worked upon: - Consider to drop the BQL dependency when reap dirty rings: then we can run the reaper thread in parallel of main thread. Needs some thoughts around the race conditions of main thread to see whether it's doable. - Consider to drop the kvmslot bitmap: logically this can be dropped with kvm dirty ring, not only for space saving, but also it's slower, and it's yet another layer linear to guest mem size which is against the whole idea of kvm dirty ring. This should make above number (of kvm dirty ring) even smaller, but probably still not as good as dirty log if workload is high. I'm not sure whether it's possible to even drop the whole ramblock dirty bitmap when kvm enabled then we remove all the bitmap caches (I guess VGA would still need a very small one if it wants, or just do the refresh unconditionally), but that's a much bigger surgery and a wild idea, but logically it should be even more efficient with the ring structure, then precopy will just work similar to postcopy that there'll be a queue of dirty pages (probably except the 1st round of precopy). I'll append them into the cover letter of next version too.