On 20/10/2018 05:15, Junaid Shahid wrote: > However, this new mode does incur additional time in the get_dirty_log > IOCTL because of having to scan all the SPTEs for dirtiness, which can > in some cases, result in slightly elongating the blackout time. Two > strategies can be used to mitigate that. The first, included in this > series, is to have the ability to dynamically switch between D Bit based > and write protection based dirty logging modes. Then the mode could be > switched to write protection just before entering blackout, thus incurring > less time in the get_dirty_log IOCTL during blackout, while keeping the > reduced overhead of the D Bit based mode during brownout. Secondly, the > scan in the get_dirty_log IOCTL can be parallelized, which results in that > IOCTL becoming faster for all modes. That patch series will be sent out > separately a little later. I should be really close now to sending the dirty page ring buffer series, which I think is incompatible with PTE-based dirty logging and basically removes the cost of ioctls almost entirely, making them O(#dirtied_pages). Do you think it's worthwhile to have PTE-based dirty logging on !PML hosts, even at the cost of disabling the ring buffer? The reduced performance for write protection is relatively small, less than 10%, and without the ring buffer we sometimes have to resort to vCPU stunning because the VM is _still_ running too fast for migration to converge. Paolo
