Opportunistic Memory Reclaim Support Proposal Updated For The Linux Kernel

[phoronix]

Phoronix: Opportunistic Memory Reclaim Support Proposal Updated For The Linux Kernel

Canonical kernel engineer Andrea Righi has sent out an updated patch series implementing opportunistic memory reclaim support as a way of forcing the kernel to attempt to reclaim system memory...

http://www.phoronix.com/scan.php?page=news_item&px=Opportunistic-Memory-Reclaim-v2

[andrei_me]

The thread in the mailing list uses hibernate as a benchmark for spot instances, can this work be combined with recent post [1] about hibernate IO stuff?
As this patches deals with memory before hibernate, and the referred post deals with IO while hibernating

[1] https://www.phoronix.com/scan.php?pa...ster-Hibernate

[arcivanov]

I hope implementation is a hint, cause even in Java triggering manual GC trigger happening too often caused system performance to grind to a halt. And with the same for kernel it can easily become a source of DoS attack unless kernel treats the userland request as a hint only but otherwise rate limits and ignores if requested too often.

[arQon]

OOM is something the Linux ecosystem still hasn't got even CLOSE to "right" still. (It's a hard problem, but it's far from an insoluble one. Funnily enough, cgroups actually provides nearly all the infrastructure needed - it's just not used at all well unless you're building a custom distro. Even just good use of *ulimit* would keep most critical outcomes from happening).

While it would be fair to call this a clumsy hack, anything that helps even the tiniest amount is going to be a win for some users. Even if this ends up being done so badly that it just stalls the machine for 9s out of every 10, it would probably still be an improvement in the typical death spiral cases.