On Wed, 6 Nov 2013 03:42:20 -0500 (EST) Jerome Marchand <jmarchan@xxxxxxxxxx> wrote: > > > ----- Original Message ----- > > From: "Andrew Morton" <akpm@xxxxxxxxxxxxxxxxxxxx> > > To: "Jerome Marchand" <jmarchan@xxxxxxxxxx> > > Cc: linux-mm@xxxxxxxxx, linux-kernel@xxxxxxxxxxxxxxx, "dave hansen" <dave.hansen@xxxxxxxxx> > > Sent: Wednesday, November 6, 2013 12:53:19 AM > > Subject: Re: [PATCH v4 2/2] mm: allow to set overcommit ratio more precisely > > > > On Fri, 18 Oct 2013 14:56:59 +0200 Jerome Marchand <jmarchan@xxxxxxxxxx> > > wrote: > > > > > Some applications that run on HPC clusters are designed around the > > > availability of RAM and the overcommit ratio is fine tuned to get the > > > maximum usage of memory without swapping. With growing memory, the 1% > > > of all RAM grain provided by overcommit_ratio has become too coarse > > > for these workload (on a 2TB machine it represents no less than > > > 20GB). > > > > > > This patch adds the new overcommit_ratio_ppm sysctl variable that > > > allow to set overcommit ratio with a part per million precision. > > > The old overcommit_ratio variable can still be used to set and read > > > the ratio with a 1% precision. That way, overcommit_ratio interface > > > isn't broken in any way that I can imagine. > > > > The way we've permanently squished this mistake in the past is to > > switch to "bytes". See /proc/sys/vm/*bytes. > > > > Would that approach work in this case? > > > > That was my first version of this patch (actually "kbytes" to avoid > overflow). > Dave raised the issue that it silently breaks the user interface: > overcommit_ratio is zero while the system behaves differently. I don't understand that at all. We keep overcommit_ratio as-is, with the same default values and add a different way of altering it. That should be back-compatible? -- 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/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>