On Tue, Mar 1, 2011 at 7:21 AM, Andrea Arcangeli <aarcange@xxxxxxxxxx> wrote: > This is important to apply in 2.6.38. The imporoved > compaction-in-kswapd logic worked much better then the upstream one, > but performance was still a little better with no compaction in > kswapd. This is also somewhat saver as it removes a feature (that is > hurting performance a bit) instead of improving it. We used a network > benchmark. This is also confirmed by Arthur on lkml using a different > multimedia workload and checking kswapd CPU utilization. Could you provide the result of benchmark and input from others in description? Sorry for bothering you but I think you get the data. It helps someone in future very much to know why we determined to remove the feature at that time and they should do what kinds of experiment to prove it has a benefit to add compaction in kswapd again. -- Kind regards, Minchan Kim -- 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>