> -----Original Message----- > From: ext Anton Vorontsov [mailto:anton.vorontsov@xxxxxxxxxx] > Sent: 08 June, 2012 15:14 > To: Moiseichuk Leonid (Nokia-MP/Espoo) ... > Hm. I would expect that avg value for meminfo will be much worse than > vmstat (meminfo grabs some locks). > > OK, if we consider 100ms interval, then this would be like 0.1% overhead? > Not great, but still better than memcg: > > http://lkml.org/lkml/2011/12/21/487 That is difficult to win over memcg :) But in comparison to one syscall like read() for small structure for particular device the generation of meminfo is about 1000x times more expensive. > So, I guess the right approach would be to find ways to not depend on > frequent vm_stat updates (and thus reads). Agree. > userland deferred timers (and infrequent reads from vmstat) + "userland vm > pressure notifications" looks promising for the userland solution. ��.n������g��� �a��� �&ޖ)���)��h���&������梷���� �Ǟ�m������)������^�����������v���O��zf������