Re: [RFC] oom notifications via /dev/oom_notify

[Balbir Singh <balbir@linux.vnet.ibm.com>]

Badari wrote:
> Rik van Riel wrote:
>> On Tue, 30 Oct 2007 13:59:28 -0800
>> Badari Pulavarty <pbadari@us.ibm.com> wrote:
>>
>>  
>>> Interesting.. Our database folks wanted some kind of notification when
>>> there is memory pressure and we are about to kill the biggest consumer
>>> (in most cases, the most useful application :(). What actually they
>>> want is a way to get notified, so that they can shrink their memory
>>> footprint in response. Just notifying before OOM may not help, since
>>> they don't have time to react. How does this notification help ? Are
>>> they supposed to monitor swapping activity and decide ?
>>>     
>>
>> Marcelo's code monitors swapping activity and will let userspace
>> programs (that poll/select the device node) know when they should
>> shrink their memory footprint.
>>
>> This is not "OOM" in the sense of "no more memory or swap", but
>> in the sense of "we're low on memory - if you don't free something
>> we'll slow you down by swapping stuff".
>>
>>   
> I think having this kind of OOM notification is a decent start. But any
> applications that
> wants to know notifications, would be more interested if kernel is
> swapping out any of
> its data, than overall system swapping events. I guess, making it
> per-process or per-cgroup
> may be logical extension. I am not sure if its really practical , though...
> 

Badari,

We have information of swapping per process in taskstats. We have
swap io count and delay. This can be easily extended to per cgroup.
It does not have the finesse of this patch, but it might meet basic
needs.

> Thanks,
> Badari


-- 
	Warm Regards,
	Balbir Singh
	Linux Technology Center
	IBM, ISTL

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>