Differences between current version and previous revision of OomKiller.
Other diffs: Previous Major Revision, Previous Author, or view the Annotated Edit History
| Newer page: | version 2 | Last edited on Saturday, November 11, 2006 6:52:00 pm | by LawrenceDoliveiro | |
| Older page: | version 1 | Last edited on Friday, May 5, 2006 10:51:49 pm | by LawrenceDoliveiro | Revert |
@@ -1,9 +1,9 @@
__~OomKiller__ -- Out-Of-Memory Process Killer
By default, the Linux kernel is configured to never say no when application processes ask for more memory. The assumption is that the applications will not actually use all the memory they ask for--this is called ''overcommitting'' memory. Hotels and airlines do the same thing when accepting bookings: the assumption is that not everybody who makes a booking will actually turn up to take their room/flight.
-However, every now and then, the assumption is wrong--you have an overbooking, and somebody has to get bumped. When Linux runs out of memory, it starts killing processes in order to free some up. Of course, the processes are chosen according to a ''heuristic'' (which is a technical term meaning "you can't please eveyone
"), and so invariably the kernel is going to kill something you consider important, thereby leading to much wailing and gnashing of teeth.
+However, every now and then, the assumption is wrong--you have an overbooking, and somebody has to get bumped. When Linux runs out of memory, it starts killing processes in order to free some up. Of course, the processes are chosen according to a ''heuristic'' (which is a technical term meaning "you can't please everyone
"), and so invariably the kernel is going to kill something you consider important, thereby leading to much wailing and gnashing of teeth.
This default behaviour can be changed via the sysctl(8) parameter <tt>vm.overcommit_memory</tt>.
Good explanation at [LWN|http://lwn.net/Articles/104179/].
