Am 15.03.2017 um 04:37 hat Lidong Chen geschrieben: > Increase bmds->cur_dirty after submit io, so reduce the frequency > involve into blk_drain, and improve the performance obviously > when block migration. > > The performance test result of this patch: > > During the block dirty save phase, this patch improve guest os IOPS > from 4.0K to 9.5K. and improve the migration speed from > 505856 rsec/s to 855756 rsec/s. > > Signed-off-by: Lidong Chen <jemmy858...@gmail.com> > --- > migration/block.c | 3 +++ > 1 file changed, 3 insertions(+) > > diff --git a/migration/block.c b/migration/block.c > index 6741228..7734ff7 100644 > --- a/migration/block.c > +++ b/migration/block.c > @@ -576,6 +576,9 @@ static int mig_save_device_dirty(QEMUFile *f, > BlkMigDevState *bmds, > } > > bdrv_reset_dirty_bitmap(bmds->dirty_bitmap, sector, nr_sectors); > + sector += nr_sectors;
sector += BDRV_SECTORS_PER_DIRTY_CHUNK would be acceptable here... > + bmds->cur_dirty = sector; > + > break; > } > sector += BDRV_SECTORS_PER_DIRTY_CHUNK; > bmds->cur_dirty = sector; > } ...which would make it the same code as the rest of the loop. It could be nicer to reuse the same code and just set a bool when a request is submitted so that the loop terminates after that. But that's just a matter of taste, so I'll still give: Reviewed-by: Kevin Wolf <kw...@redhat.com>