LSFMM: Improving the out-of-memory killer

dcymbala — Sun, 05 May 2013 00:15:20 +0000

It seems like there are a few standard types of processes that could implement a policy with the kernel which would allow the process to be shut down until conditions improve.

Server processes might provide a "stub" of code that allows active ports to remain active but return a standard protocol response to requests rather than denying requests altogether.

It might also be useful for processes to be swapped out completely, and reactivated only when the system decides it can do so. It would then also redirect the active ports off the "stub" and back to the main process.

Other kinds of processes are harder to decide on a policy. This "stub" idea would probably not work with graphical apps, but there may be other classes of processes where it would work.

I was also wondering if it would be possible to set up a system where the kernel would only run processes that have OOM policies set up. Then there would be no "rogue" processes in an otherwise harmonious mix of apps with policies.

LWN: Comments on "LSFMM: Improving the out-of-memory killer"

LSFMM: Improving the out-of-memory killer