Messages in this thread | | | Date | Mon, 3 Aug 2009 09:17:01 -0700 | Subject | Re: [patch -mm v2] mm: introduce oom_adj_child | From | Paul Menage <> |
| |
On Sat, Aug 1, 2009 at 1:26 PM, David Rientjes<rientjes@google.com> wrote: > > It's more likely than not that applications were probably written to the > way the documentation described the two files: that is, adjust > /proc/pid/oom_score by tuning /proc/pid/oom_adj
I'd actually be pretty surprised if anyone was really doing that - don't forget that the oom_score is something that varies dynamically depending on things like the VM size of the process, its running time, the VM sizes of its children, etc. So tuning oom_adj based on oom_score will be rapidly out of date. AFAIK, oom_score was added initially as a way to debug the OOM killer, and oom_adj was added later as an additional knob. My suspicion is that any automated users of oom_adj are working along the lines of
http://www.google.com/codesearch/p?hl=en&sa=N&cd=4&ct=rc#X7-oBZ_RyNM/src/server/memory/base/oommanager.cpp&q=oom_score
which just uses the values -16, 0 or 15, depending on whether the process is critical, important or expendable.
Paul
| |