Re: [PATCH 3/4] oom: oom-killer don't use permillage of system-raminternally

[KOSAKI Motohiro]

(2011/05/11 8:40), David Rientjes wrote:
> On Tue, 10 May 2011, KOSAKI Motohiro wrote:
>
> > CAI Qian reported his kernel did hang-up if he ran fork intensive
> > workload and then invoke oom-killer.
> >
> > The problem is, Current oom calculation uses 0-1000 normalized value
> > (The unit is a permillage of system-ram). Its low precision make
> > a lot of same oom score. IOW, in his case, all processes have<1
> > oom score and internal integral calculation round it to 1. Thus
> > oom-killer kill ineligible process. This regression is caused by
> > commit a63d83f427 (oom: badness heuristic rewrite).
> >
> > The solution is, the internal calculation just use number of pages
> > instead of permillage of system-ram. And convert it to permillage
> > value at displaying time.
> >
> > This patch doesn't change any ABI (included  /proc/<pid>/oom_score_adj)
> > even though current logic has a lot of my dislike thing.
>
> s/permillage/proportion/
>
> This is unacceptable, it does not allow users to tune oom_score_adj
> appropriately based on the scores exported by /proc/pid/oom_score to
> discount an amount of RAM from a thread's memory usage in systemwide,
> memory controller, cpuset, or mempolicy contexts.  This is only possible
> because the oom score is normalized.

You misunderstand the code. The patch doesn't change oom_score.
The patch change fs/proc too.

> What would be acceptable would be to increase the granularity of the score
> to 10000 or 100000 to differentiate between threads using 0.01% or 0.001%
> of RAM from each other, respectively.  The range of oom_score_adj would
> remain the same, however, and be multiplied by 10 or 100, respectively,
> when factored into the badness score baseline.  I don't believe userspace
> cares to differentiate between more than 0.1% of available memory.

Currently, SGI buy 16TB memory. 16TB x 0.1% = 1.6GB. I don't think your
fork bomb process use bigger than 1.6GB. Thus your patch is unacceptable.

So, please read the code again. or run it.

> The other issue that this patch addresses is the bonus given to root
> processes.  I agree that if a root process is using 4% of RAM that it
> should not be equal to all other threads using 1%.  I do believe that a
> root process using 60% of RAM should be equal priority to a thread using
> 57%, however.  Perhaps a compromise would be to give root processes a
> bonus of 1% for every 30% of RAM they consume?

I think you are talking about patch [4/4], right? patch [3/4] and [4/4]
are attacking another issue. big machine issue and root user issue.


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/