In the last month, I added round-robin scan of numa nodes at hittling limit, and wrote "a better algorithm is needed." Here is update. Because some of patches are bugfixes, I may cut out them as independent patch. Pathc 6-7/7 implements a selection logic. == Tested on 8cpu/24GB system, which has 2 nodes. limit memory to 300MB and run httpd under it. httpd's working set is 4096files/600MB. Then, do 40960 access by apache-bench. and see how memory reclaim costs. Because a thread of httpd doesn't consume cpu much, the number of working threads are not balanced between numa nodes and file caches will be not balanced. [round-robin] [kamezawa@bluextal ~]$ cat /cgroup/memory/test/memory.scan_stat scanned_pages_by_limit 550740 freed_pages_by_limit 206473 elapsed_ns_by_limit 9485418834 [After patch] scanned_pages_by_limit 521520 freed_pages_by_limit 199330 elapsed_ns_by_limit 7904913234 I can see elapsed time is decreased. Test on big machine is welcomed. Thanks, -Kame -- 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>