On Tue, Jan 31, 2017 at 03:58:16PM +1100, Anton Blanchard wrote: >Hi, > > >> Anton, I think the behaviour looks good. Actually, it's not very >> relevant to the issue addressed by the patch. I will reply to >> Michael's reply about the reason. There are two nodes in your system >> and the memory is expected to be allocated from node-0. If node-0 >> doesn't have enough free memory, the allocater switches to node-1. It >> means we need more stress. > >Did you try setting zone_reclaim_mode? Surely we should reclaim local >clean pagecache if enabled? >
In last experiment, I didn't enable zone_reclaim_mode. After changed it to 0x2 (RECLAIM_WRITE), the local pagecache isn't reclaimed from node-0 as we observed before. root@P83-p1:~# cat /proc/sys/vm/zone_reclaim_mode 2 root@P83-p1:~# sync root@P83-p1:~# echo 3 > /proc/sys/vm/drop_caches root@P83-p1:~# taskset -c 0 cat /tmp/file.8G > /dev/null root@P83-p1:~# grep FilePages /sys/devices/system/node/node0/meminfo Node 0 FilePages: 8497920 kB root@P83-p1:~# taskset -c 0 ./alloc 68719476736 root@P83-p1:~# grep FilePages /sys/devices/system/node/node0/meminfo Node 0 FilePages: 8497920 kB With the patch applied, the local pagecache is reclaimed: root@P83-p1:~# cat /proc/sys/vm/zone_reclaim_mode 2 root@P83-p1:~# sync root@P83-p1:~# echo 3 > /proc/sys/vm/drop_caches root@P83-p1:~# taskset -c 0 cat /tmp/file.8G > /dev/null root@P83-p1:~# grep FilePages /sys/devices/system/node/node0/meminfo Node 0 FilePages: 8441472 kB root@P83-p1:~# taskset -c 0 ./alloc 68719476736 root@P83-p1:~# grep FilePages /sys/devices/system/node/node0/meminfo Node 0 FilePages: 712960 kB Thanks, Gavin
