Re: [PATCH 2/2] vmscan: kill shrink_all_zones()

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Friday 09 October 2009, KOSAKI Motohiro wrote:
> shrink_all_zone() was introduced by commit d6277db4ab (swsusp: rework
> memory shrinker) for hibernate performance improvement. and sc.swap_cluster_max
> was introduced by commit a06fe4d307 (Speed freeing memory for suspend).
> 
> commit a06fe4d307 said
> 
>     Without the patch:
>     Freed  14600 pages in  1749 jiffies = 32.61 MB/s (Anomolous!)
>     Freed  88563 pages in 14719 jiffies = 23.50 MB/s
>     Freed 205734 pages in 32389 jiffies = 24.81 MB/s
> 
>     With the patch:
>     Freed  68252 pages in   496 jiffies = 537.52 MB/s
>     Freed 116464 pages in   569 jiffies = 798.54 MB/s
>     Freed 209699 pages in   705 jiffies = 1161.89 MB/s
> 
> At that time, their patch was pretty worth. However, Modern Hardware
> trend and recent VM improvement broke its worth. From several reason,
> I think we should remove shrink_all_zones() at all.
> 
> detail:
> 
> 1) Old days, shrink_zone()'s slowness was mainly caused by stupid congestion_wait()
>    at no i/o congestion.
>    but current shrink_zone() is sane, not slow.
> 
> 2) shrink_all_zone() try to shrink all pages at a time. but it doesn't works
>    fine on numa system.
>    example)
>      System has 4GB memory and each node have 2GB. and hibernate need 1GB.
> 
>      optimal)
> 	steal 500MB from each node.
>      shrink_all_zones)
> 	steal 1GB from node-0.
> 
>    Oh, Cache balancing was broke ;)
>    Unfortunately, Desktop system moved ahead NUMA.
>    (Side note, if hibernate require 2GB, shrink_all_zones() never success)
> 
> 3) if the node has several I/O flighting pages, shrink_all_zones() makes
>    pretty bad result.
> 
>    schenario) hibernate need 1GB
> 
>    1) shrink_all_zones() try to reclaim 1GB from Node-0
>    2) but it only reclaimed 990MB
>    3) stupidly, shrink_all_zones() try to reclaim 1GB from Node-1
>    4) it reclaimed 990MB
> 
>    Oh, well. it reclaimed twice much than required.
>    In the other hand, current shrink_zone() has sane baling out logic.
>    then, it doesn't make overkill reclaim. then, we lost shrink_zones()'s risk.
> 
> 4) SplitLRU VM always keep active/inactive ratio very carefully. inactive list only
>    shrinking break its assumption. it makes unnecessary OOM risk. it obviously suboptimal.
> 
>   Throuput comparision
>   ==============================================
>   old		2192.10 MB/s
>   new		2222.22 MB/s
> 
>   ok, it's almost same throuput.
> 
> Cc: Rafael J. Wysocki <rjw@sisk.pl>
> Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>

I have no objections to any of the two patches, but I think we may want to drop
shrink_all_memory() altogether.  Everything should work without it and the
reason I didn't remove it was because I saw a performance regression on one
system without it.  It may not be worth keeping it, though.

Have you done any tests with shrink_all_memory() removed?

Rafael


> ---
>  mm/vmscan.c |   75 +++++++++++++---------------------------------------------
>  1 files changed, 17 insertions(+), 58 deletions(-)
> 
> diff --git a/mm/vmscan.c b/mm/vmscan.c
> index 80e727d..9f28166 100644
> --- a/mm/vmscan.c
> +++ b/mm/vmscan.c
> @@ -2130,51 +2130,6 @@ unsigned long global_lru_pages(void)
>  
>  #ifdef CONFIG_HIBERNATION
>  /*
> - * Helper function for shrink_all_memory().  Tries to reclaim 'nr_pages' pages
> - * from LRU lists system-wide, for given pass and priority.
> - *
> - * For pass > 3 we also try to shrink the LRU lists that contain a few pages
> - */
> -static void shrink_all_zones(unsigned long nr_pages, int prio,
> -				      int pass, struct scan_control *sc)
> -{
> -	struct zone *zone;
> -	unsigned long nr_reclaimed = 0;
> -
> -	for_each_populated_zone(zone) {
> -		enum lru_list l;
> -
> -		if (zone_is_all_unreclaimable(zone) && prio != DEF_PRIORITY)
> -			continue;
> -
> -		for_each_evictable_lru(l) {
> -			enum zone_stat_item ls = NR_LRU_BASE + l;
> -			unsigned long lru_pages = zone_page_state(zone, ls);
> -
> -			/* For pass = 0, we don't shrink the active list */
> -			if (pass == 0 && (l == LRU_ACTIVE_ANON ||
> -						l == LRU_ACTIVE_FILE))
> -				continue;
> -
> -			zone->lru[l].nr_saved_scan += (lru_pages >> prio) + 1;
> -			if (zone->lru[l].nr_saved_scan >= nr_pages || pass > 3) {
> -				unsigned long nr_to_scan;
> -
> -				zone->lru[l].nr_saved_scan = 0;
> -				nr_to_scan = min(nr_pages, lru_pages);
> -				nr_reclaimed += shrink_list(l, nr_to_scan, zone,
> -								sc, prio);
> -				if (nr_reclaimed >= nr_pages) {
> -					sc->nr_reclaimed += nr_reclaimed;
> -					return;
> -				}
> -			}
> -		}
> -	}
> -	sc->nr_reclaimed += nr_reclaimed;
> -}
> -
> -/*
>   * Try to free `nr_pages' of memory, system-wide, and return the number of
>   * freed pages.
>   *
> @@ -2188,12 +2143,18 @@ unsigned long shrink_all_memory(unsigned long nr_pages)
>  	int pass;
>  	struct reclaim_state reclaim_state;
>  	struct scan_control sc = {
> -		.gfp_mask = GFP_KERNEL,
> +		.gfp_mask = GFP_HIGHUSER_MOVABLE,
> +		.may_swap = 0,
>  		.may_unmap = 0,
>  		.may_writepage = 1,
> +		.swap_cluster_max = SWAP_CLUSTER_MAX,
> +		.nr_to_reclaim = nr_pages,
> +		.swappiness = vm_swappiness,
> +		.order = 0,
>  		.isolate_pages = isolate_pages_global,
> -		.nr_reclaimed = 0,
>  	};
> +	struct zonelist * zonelist = node_zonelist(first_online_node,
> +						   GFP_HIGHUSER_MOVABLE);
>  
>  	current->reclaim_state = &reclaim_state;
>  
> @@ -2215,9 +2176,9 @@ unsigned long shrink_all_memory(unsigned long nr_pages)
>  
>  	/*
>  	 * We try to shrink LRUs in 5 passes:
> -	 * 0 = Reclaim from inactive_list only
> -	 * 1 = Reclaim from active list but don't reclaim mapped
> -	 * 2 = 2nd pass of type 1
> +	 * 0 = Reclaim unmapped pages
> +	 * 1 = 2nd pass of type 0
> +	 * 2 = 3rd pass of type 0
>  	 * 3 = Reclaim mapped (normal reclaim)
>  	 * 4 = 2nd pass of type 3
>  	 */
> @@ -2225,15 +2186,15 @@ unsigned long shrink_all_memory(unsigned long nr_pages)
>  		int prio;
>  
>  		/* Force reclaiming mapped pages in the passes #3 and #4 */
> -		if (pass > 2)
> +		if (pass > 2) {
>  			sc.may_unmap = 1;
> +			sc.may_swap = 1;
> +		}
>  
>  		for (prio = DEF_PRIORITY; prio >= 0; prio--) {
> -			unsigned long nr_to_scan = nr_pages - sc.nr_reclaimed;
> -
>  			sc.nr_scanned = 0;
> -			sc.swap_cluster_max = nr_to_scan;
> -			shrink_all_zones(nr_to_scan, prio, pass, &sc);
> +
> +			shrink_zones(prio, zonelist, &sc);
>  			if (sc.nr_reclaimed >= nr_pages)
>  				goto out;
>  
> @@ -2243,10 +2204,8 @@ unsigned long shrink_all_memory(unsigned long nr_pages)
>  			sc.nr_reclaimed += reclaim_state.reclaimed_slab;
>  			if (sc.nr_reclaimed >= nr_pages)
>  				goto out;
> -
> -			if (sc.nr_scanned && prio < DEF_PRIORITY - 2)
> -				congestion_wait(BLK_RW_ASYNC, HZ / 10);
>  		}
> +		congestion_wait(BLK_RW_ASYNC, HZ / 10);
>  	}
>  
>  	/*
>