Re: Performance of partial object-based rmap

[Martin J. Bligh]

> The performance delta between 2.5.62-mjb1 and 2.5.62-mjb2 is caused
> by the partial object-based rmap patch (written by Dave McCracken).
> I expect this patch to have an increasing impact on workloads with
> more processes, and it should give a substantial space saving as 
> well as a performance increase. Results from 16x NUMA-Q system ... 
> 
> Profile comparison:
> 
> before
> 	15525 page_remove_rmap
> 	6415 page_add_rmap
> 
> after
> 	2055 page_add_rmap
> 	1983 page_remove_rmap

Did some space consumption comparisons on make -j 256:

before:
	24116 pte_chain objects in slab cache
after:
	716 pte_chain objects in slab cache

The vast majority of anonymous pages (for which we're using the non
object based method) are singletons, and hence use pte_direct ...
hence the massive space reduction.

M.