Re: [RFC PATCH v1 13/13] mm: splice local lists onto the front of the LRU

[Daniel Jordan]

On 02/02/2018 12:21 AM, Aaron Lu wrote:
> On Wed, Jan 31, 2018 at 06:04:13PM -0500, daniel.m.jordan@xxxxxxxxxx wrote:
> > Now that release_pages is scaling better with concurrent removals from
> > the LRU, the performance results (included below) showed increased
> > contention on lru_lock in the add-to-LRU path.
> >
> > To alleviate some of this contention, do more work outside the LRU lock.
> > Prepare a local list of pages to be spliced onto the front of the LRU,
> > including setting PageLRU in each page, before taking lru_lock.  Since
> > other threads use this page flag in certain checks outside lru_lock,
> > ensure each page's LRU links have been properly initialized before
> > setting the flag, and use memory barriers accordingly.
> >
> > Performance Results
> >
> > This is a will-it-scale run of page_fault1 using 4 different kernels.
> >
> >              kernel     kern #
> >
> >            4.15-rc2          1
> >    large-zone-batch          2
> >       lru-lock-base          3
> >     lru-lock-splice          4
> >
> > Each kernel builds on the last.  The first is a baseline, the second
> > makes zone->lock more scalable by increasing an order-0 per-cpu
> > pagelist's 'batch' and 'high' values to 310 and 1860 respectively
>
> Since the purpose of the patchset is to optimize lru_lock, you may
> consider adjusting pcp->high to be >= 32768(page_fault1's test size is
> 128M = 32768 pages). That should eliminate zone->lock contention
> entirely.

Interesting, hadn't thought about taking zone->lock completely out of 
the equation.  Will try this next time I test this series.


While we're on this topic, it does seem from the performance of kernel 
#2, and the numbers Aaron posted in a previous thread[*], that the 
default 'batch' and 'high' values should be bigger on large systems.

The code to control these two values last changed in 2005[**], so we hit 
the largest values with just a 512M zone:

   zone       4k_pages  batch   high  high/4k_pages
    64M         16,384      3     18       0.10986%
   128M         32,768      7     42       0.12817%
   256M         65,536     15     90       0.13733%
   512M        131,072     31    186       0.14191%
     1G        262,144     31    186       0.07095%
     2G        524,288     31    186       0.03548%
     4G      1,048,576     31    186       0.01774%
     8G      2,097,152     31    186       0.00887%
    16G      4,194,304     31    186       0.00443%
    32G      8,388,608     31    186       0.00222%
    64G     16,777,216     31    186       0.00111%
   128G     33,554,432     31    186       0.00055%
   256G     67,108,864     31    186       0.00028%
   512G    134,217,728     31    186       0.00014%
  1024G    268,435,456     31    186       0.00007%


[*] https://marc.info/?l=linux-netdev&m=150572010919327
[**] ba56e91c9401 ("[PATCH] mm: page_alloc: increase size of per-cpu-pages")

> > (courtesy of Aaron Lu's patch), the third scales lru_lock without
> > splicing pages (the previous patch in this series), and the fourth adds
> > page splicing (this patch).
> >
> > N tasks mmap, fault, and munmap anonymous pages in a loop until the test
> > time has elapsed.