Michal Hocko <mhocko@xxxxxxxxxx> writes:
> On Tue 03-03-20 19:49:53, Huang, Ying wrote:
>> Michal Hocko <mhocko@xxxxxxxxxx> writes:
>>
>> > On Tue 03-03-20 16:47:54, Huang, Ying wrote:
>> >> Michal Hocko <mhocko@xxxxxxxxxx> writes:
>> >>
>> >> > On Tue 03-03-20 09:51:56, Huang, Ying wrote:
>> >> >> Mel Gorman <mgorman@xxxxxxx> writes:
>> >> >> > On Mon, Mar 02, 2020 at 07:23:12PM +0800, Huang, Ying wrote:
>> >> >> >> If some applications cannot tolerate the latency incurred by the memory
>> >> >> >> allocation and zeroing. Then we cannot discard instead of migrate
>> >> >> >> always. While in some situations, less memory pressure can help. So
>> >> >> >> it's better to let the administrator and the application choose the
>> >> >> >> right behavior in the specific situation?
>> >> >> >>
>> >> >> >
>> >> >> > Is there an application you have in mind that benefits from discarding
>> >> >> > MADV_FREE pages instead of migrating them?
>> >> >> >
>> >> >> > Allowing the administrator or application to tune this would be very
>> >> >> > problematic. An application would require an update to the system call
>> >> >> > to take advantage of it and then detect if the running kernel supports
>> >> >> > it. An administrator would have to detect that MADV_FREE pages are being
>> >> >> > prematurely discarded leading to a slowdown and that is hard to detect.
>> >> >> > It could be inferred from monitoring compaction stats and checking
>> >> >> > if compaction activity is correlated with higher minor faults in the
>> >> >> > target application. Proving the correlation would require using the perf
>> >> >> > software event PERF_COUNT_SW_PAGE_FAULTS_MIN and matching the addresses
>> >> >> > to MADV_FREE regions that were freed prematurely. That is not an obvious
>> >> >> > debugging step to take when an application detects latency spikes.
>> >> >> >
>> >> >> > Now, you could add a counter specifically for MADV_FREE pages freed for
>> >> >> > reasons other than memory pressure and hope the administrator knows about
>> >> >> > the counter and what it means. That type of knowledge could take a long
>> >> >> > time to spread so it's really very important that there is evidence of
>> >> >> > an application that suffers due to the current MADV_FREE and migration
>> >> >> > behaviour.
>> >> >>
>> >> >> OK. I understand that this patchset isn't a universal win, so we need
>> >> >> some way to justify it. I will try to find some application for that.
>> >> >>
>> >> >> Another thought, as proposed by David Hildenbrand, it's may be a
>> >> >> universal win to discard clean MADV_FREE pages when migrating if there are
>> >> >> already memory pressure on the target node. For example, if the free
>> >> >> memory on the target node is lower than high watermark?
>> >> >
>> >> > This is already happening because if the target node is short on memory
>> >> > it will start to reclaim and if MADV_FREE pages are at the tail of
>> >> > inactive file LRU list then they will be dropped. Please note how that
>> >> > follows proper aging and doesn't introduce any special casing. Really
>> >> > MADV_FREE is an inactive cache for anonymous memory and we treat it like
>> >> > inactive page cache. This is not carved in stone of course but it really
>> >> > requires very good justification to change.
>> >>
>> >> If my understanding were correct, the newly migrated clean MADV_FREE
>> >> pages will be put at the head of inactive file LRU list instead of the
>> >> tail. So it's possible that some useful file cache pages will be
>> >> reclaimed.
>> >
>> > This is the case also when you migrate other pages, right? We simply
>> > cannot preserve the aging.
>>
>> So you consider the priority of the clean MADV_FREE pages is same as
>> that of page cache pages?
>
> This is how MADV_FREE has been implemented, yes. See f7ad2a6cb9f7 ("mm:
> move MADV_FREE pages into LRU_INACTIVE_FILE list") for the
> justification.
Thanks for information. It's really helpful!
>> Because the penalty difference is so large, I
>> think it may be a good idea to always put clean MADV_FREE pages at the
>> tail of the inactive file LRU list?
>
> You are again making assumptions without giving any actual real
> examples. Reconstructing MADV_FREE pages cost can differ a lot.
In which situation the cost to reconstruct MADV_FREE pages can be higher
than the cost to allocate file cache page and read from disk? Heavy
contention on mmap_sem?
> This really depends on the specific usecase. Moving pages to the tail
> of LRU would make them the primary candidate for the reclaim with a
> strange LIFO semantic. Adding them to the head might be not the
> universal win but it will at least provide a reasonable FIFO
> semantic. I also find it much more easier to reason about MADV_FREE as
> an inactive cache.
Yes. FIFO is more reasonable than LIFO.
Best Regards,
Huang, Ying
