On (23/03/15 08:58), Christoph Hellwig wrote: > > I think this simply accounts for the I/Os now that were skipped > when using the bdev_read/write path before. Oh, that would explain it. Otherwise I was slightly surprised (in a good way) and puzzled.
