On Fri, May 06, 2011 at 12:37:08AM +0800, Jan Kara wrote: > On Wed 04-05-11 15:39:31, Wu Fengguang wrote: > > To help understand the behavior change, I wrote the writeback_queue_io > > trace event, and found very different patterns between > > - vanilla kernel > > - this patchset plus the sync livelock fixes > > > > Basically the vanilla kernel each time pulls a random number of inodes > > from b_dirty, while the patched kernel tends to pull a fixed number of > > inodes (enqueue=1031) from b_dirty. The new behavior is very interesting... > This regularity is really strange. Did you have a chance to look more into > it? I find it highly unlikely that there would be exactly 1031 dirty inodes > in b_dirty list every time you call move_expired_inodes()... Yeah that's the weird point. The other things I noticed are more regular "flusher - dd - flusher - dd - ..." writeout patterns after the patches. In vanilla kernel it behaves more randomly and there are many balance_dirty_pages() IOs from tar. I'll try to collect more traces in ext4 tomorrow. Sorry it's too late for me now. Thanks, Fengguang -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxxx For more info on Linux MM, see: http://www.linux-mm.org/ . Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/ Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>