Re: [PATCH] shrink core hashes on small systems

[Matt Mackall]

On Mon, Apr 05, 2004 at 02:02:23PM -0700, Andrew Morton wrote:
> Matt Mackall <mpm@xxxxxxxxxxx> wrote:
> >
> > Shrink hashes on small systems
> > 
> > Tweak vfs_caches_init logic so that hashes don't start growing as
> > quickly on small systems.
> > 
> > -	vfs_caches_init(num_physpages);
> > +	/* Treat machines smaller than 6M as having 2M of memory
> > +	   for hash-sizing purposes */
> > +	vfs_caches_init(max(500, (int)num_physpages-1000));
> 
> This seems rather arbitrary.  It also implicitly "knows" that
> PAGE_SIZE=4096.

Yep. I can reword it in terms of pages, if that helps. Boxes with 8k
pages tend to have larger instruction words and data structures by
virtue of being RISC/64bit/etc., so I think 1000 pages is a reasonable
number in either case.
 
> num_physpages is of course the wrong thing to use here - on small systems
> we should be accounting for memory which is pinned by kernel text, etc.
> 
> But you're going further than that.  What's the theory here?

Basically, the numfreepages approach doesn't take into account the
size of the kernel/critical userspace at all. So we assume that
anything less than 4M is already tight and that we're not yet in a
position to trade space for performance, so lets just pull that off the top.

Longer term, I think some serious thought needs to go into scaling
hash sizes across the board, but this serves my purposes on the
low-end without changing behaviour asymptotically.

-- 
Matt Mackall : http://www.selenic.com : Linux development and consulting
-
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/