On Thu, 2011-07-28 at 07:45 +0800, Andrew Morton wrote: > On Sat, 23 Jul 2011 20:49:10 +0200 > Jens Axboe <jaxboe@xxxxxxxxxxxx> wrote: > > > > I can observe the average request size changes. Before the patch, the > > > average request size is about 90k from iostat (but the variation is > > > big). With the patch, the request size is about 100k and variation is > > > small. > > > > That's a good win right there, imho. > > yup. Reduced CPU consumption on that path isn't terribly exciting IMO, > but improved request size is significant. > > Using an additional 44 bytes of stack on that path is also > significant(ly bad). But we need to fix that problem anyway. One way > we could improve things in mm/vmscan.c is to move the blk_plug into > scan_control then get the scan_control off the stack in some manner. > That's easy for kswapd: allocate one scan_control per kswapd at > startup. Doing it for direct-reclaim would be a bit trickier... unfortunately, the direct-reclaim case is what cares about stack. BTW, the scan_control can be dieted. may_unmap/may_swap/may_writepage can be a bit. swappiness < 100, so can be a char. order <= 11, can be a char. should I do it to cut the size? > And I have the usual maintainability whine. If someone comes up to > vmscan.c and sees it calling blk_start_plug(), how are they supposed to > work out why that call is there? They go look at the blk_start_plug() > definition and it is undocumented. I think we can do better than this? the block plug is a little tricky, we definitely should document it. Jens, if you don't mind, I'll add comments there. Thanks, Shaohua -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. 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>