Re: [PATCH] vmscan: fix slab vs lru balance

[Andrew Morton]

On Tue, 24 Nov 2015 15:47:21 +0300 Vladimir Davydov <vdavydov@xxxxxxxxxxxxx> wrote:

> The comment to shrink_slab states that the portion of kmem objects
> scanned by it equals the portion of lru pages scanned by shrink_zone
> over shrinker->seeks.
> 
> shrinker->seeks is supposed to be equal to the number of disk seeks
> required to recreated an object. It is usually set to DEFAULT_SEEKS (2),
> which is quite logical, because most kmem objects (e.g. dentry or inode)
> require random IO to reread (seek to read and seek back).
> 
> That said, one would expect that dcache is scanned two times less
> intensively than page cache, which sounds sane as dentries are generally
> more costly to recreate.
> 
> However, the formula for distributing memory pressure between slab and
> lru actually looks as follows (see do_shrink_slab):
> 
>                               lru_scanned
> objs_to_scan = objs_total * --------------- * 4 / shrinker->seeks
>                             lru_reclaimable
> 
> That is dcache, as well as most of other slab caches, is scanned two
> times more aggressively than page cache.
> 
> Fix this by dropping '4' from the equation above.
> 

oh geeze.  Who wrote that crap?


commit c3f4656118a78c1c294e0b4d338ac946265a822b
Author: Andrew Morton <akpm@xxxxxxxx>
Date:   Mon Dec 29 23:48:44 2003 -0800

    [PATCH] shrink_slab acounts for seeks incorrectly
    
    wli points out that shrink_slab inverts the sense of shrinker->seeks: those
    caches which require more seeks to reestablish an object are shrunk harder.
    That's wrong - they should be shrunk less.
    
    So fix that up, but scaling the result so that the patch is actually a no-op
    at this time, because all caches use DEFAULT_SEEKS (2).

diff --git a/mm/vmscan.c b/mm/vmscan.c
index b859482..f2da3c9 100644
--- a/mm/vmscan.c
+++ b/mm/vmscan.c
@@ -154,7 +154,7 @@ static int shrink_slab(long scanned, unsigned int gfp_mask)
 	list_for_each_entry(shrinker, &shrinker_list, list) {
 		unsigned long long delta;
 
-		delta = scanned * shrinker->seeks;
+		delta = 4 * (scanned / shrinker->seeks);
 		delta *= (*shrinker->shrinker)(0, gfp_mask);
 		do_div(delta, pages + 1);
 		shrinker->nr += delta;


What a pathetic changelog.

The current code may be good, it may be bad, but I'm reluctant to
change it without a solid demonstration that the result is overall
superior.

--
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/