On Thu, Apr 11, 2013 at 10:57:08PM -0400, Theodore Ts'o wrote: > On Thu, Apr 11, 2013 at 11:33:35PM +0200, Jan Kara wrote: > > I think it might be more enlightening if Mel traced which process in > > which funclion is holding the buffer lock. I suspect we'll find out that > > the flusher thread has submitted the buffer for IO as an async write and > > thus it takes a long time to complete in presence of reads which have > > higher priority. > > That's an interesting theory. If the workload is one which is very > heavy on reads and writes, that could explain the high latency. That > would explain why those of us who are using primarily SSD's are seeing > the problems, because would be reads are nice and fast. > > If that is the case, one possible solution that comes to mind would be > to mark buffer_heads that contain metadata with a flag, so that the > flusher thread can write them back at the same priority as reads. > > The only problem I can see with this hypothesis is that if this is the > explanation for what Mel and Jiri are seeing, it's something that > would have been around for a long time, and would affect ext3 as well > as ext4. That isn't quite consistent, however, with Mel's observation > that this is a probablem which has gotten worse in relatively > recently. > According to the tests I've run, multi-second stalls have been a problem for a while but never really bothered me. I'm not sure why it felt particularly bad around -rc2 or why it seems to be better now. Maybe I just had my cranky pants on. -- Mel Gorman SUSE Labs -- To unsubscribe from this list: send the line "unsubscribe linux-ext4" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
