Re: [PATCH] Radix-tree pagecache for 2.5

[Linus Torvalds]

On 30 Jan 2002, Momchil Velikov wrote:
>
> rat-7 is with 128-way radix tree branch factor, rat-4 is with 16-way.

Hmm. It appears that the 128-way one is no slower, at least.

Probably not very relevant question: What are the memory usage
implications? I love having that global big page_hash_table gone, but what
are the differences in memory usage between rat-4 and rat-7? In
particular, it _looks_ like the way the radix_node is done, it will
basically always be a factor-of-two+1 words, which sounds like the worst
possible schenario from an allocator standpoint.

		Linus