Re: [patch 1/3] slob: rework freelist handling

[Nick Piggin <npiggin@suse.de>]

On Tue, May 22, 2007 at 11:38:54PM -0700, Christoph Lameter wrote:
> On Wed, 23 May 2007, Nick Piggin wrote:
> 
> > OK, so with a 64-bit UP ppc kernel, compiled for size, and without full
> > size data structures, booting with mem=16M init=/bin/bash.
> 
> Hmmm.. Cannot do much on such a system. Try a 32 bit instead?

I could try, at some point.

  
> > After booting and mounting /proc, SLOB has 1140K free, SLUB has 748K
> > free.
> 
> The following patch may help a little bit but not much. Hmmm... In order 
> to reduce the space further we would also have to shrink all caches when 
> boot is  complete. Elimination of useless caches also would be good. 
> Do you really want to go into this deeper?

Well you asked the question what good is SLOB when we have SLUB, and the
answer, not surprisingly, still seems to be that it is better for memory
constrained environments.

I'm happy to test any patches from you. If you are able to make SLUB as
space efficient as SLOB on small systems, that would be great, and we
could talk about replacing that too. I think it would be a hefty task,
though.


> ---
>  include/linux/slub_def.h |    2 ++
>  mm/slub.c                |    4 +++-
>  2 files changed, 5 insertions(+), 1 deletion(-)
> 
> Index: slub/include/linux/slub_def.h
> ===================================================================
> --- slub.orig/include/linux/slub_def.h	2007-05-22 22:46:06.000000000 -0700
> +++ slub/include/linux/slub_def.h	2007-05-22 23:31:18.000000000 -0700
> @@ -17,7 +17,9 @@ struct kmem_cache_node {
>  	unsigned long nr_partial;
>  	atomic_long_t nr_slabs;
>  	struct list_head partial;
> +#ifdef CONFIG_SLUB_DEBUG
>  	struct list_head full;
> +#endif
>  };
>  
>  /*
> Index: slub/mm/slub.c
> ===================================================================
> --- slub.orig/mm/slub.c	2007-05-22 22:46:06.000000000 -0700
> +++ slub/mm/slub.c	2007-05-22 23:32:00.000000000 -0700
> @@ -183,7 +183,7 @@ static inline void ClearSlabDebug(struct
>   * Mininum number of partial slabs. These will be left on the partial
>   * lists even if they are empty. kmem_cache_shrink may reclaim them.
>   */
> -#define MIN_PARTIAL 2
> +#define MIN_PARTIAL 0
>  
>  /*
>   * Maximum number of desirable partial slabs.
> @@ -1792,7 +1792,9 @@ static void init_kmem_cache_node(struct 
>  	atomic_long_set(&n->nr_slabs, 0);
>  	spin_lock_init(&n->list_lock);
>  	INIT_LIST_HEAD(&n->partial);
> +#ifdef CONFIG_SLUB_DEBUG
>  	INIT_LIST_HEAD(&n->full);
> +#endif
>  }
>  
>  #ifdef CONFIG_NUMA

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>