Hi,

On Mon, Jul 21, 2014 at 01:45:07PM +0200, Pavel Machek wrote:

Ok, so you have sped it up from O(n) to O(log(n)) speed, and increased
memory requirements from O(n) to O(n * log(n)), right?

No, new memory requirements are still ~O(n), because the additional pages
needed for the index in the radix tree are outweighted by the memory
saved by the smaller size of struct rtree_node compared to struct
bm_block.

If you want to see real numbers, here is a small tool which tells you
exactly when you need more memory and how much you save with the new
memory bitmap implementation:

	http://www.zlug.org/~joro/data/mem_req.c

It will tell you that the new implementation uses at most 2 more pages
for smaller RAM sizes (2 pages because 2 memory bitmaps are allocated)
and will actually save pages for bigger RAM sizes (already 30 pages on a
1TB machine, 384 pages on a 12TB machine)

You can also modify the above tool to give you some data to plot, then
you will also SEE that there is still a linear relationship between RAM
size and memory required by the old and new memory bitmap
implementation.


	Joerg