Re: [Lse-tech] Re: Performance of partial object-based rmap

[Martin J. B. <mb...@ar...>]

>>> Could I get a larger, multiplicative differential profile?
>>> i.e. ratios of the fractions of profile hits?
>>> If you have trouble generating such I can do so myself from
>>> fuller profile results.
> 
> On Wed, Feb 19, 2003 at 09:14:10PM -0800, Martin J. Bligh wrote:
>> before:
>> 	187256 total
>> after:
>> 	170196 total
> 
> Well, I was trying to get an idea of what got slower to compensate
> for the improvement in page_(add|remove)_rmap() times.

diffprofile:

1514 default_idle
921 .text.lock.file_table
145 vm_enough_memory
131 release_pages
93 file_move
68 __copy_to_user_ll
62 fd_install
60 buffered_rmqueue
58 vma_merge
52 __fput
...
-65 kmem_cache_alloc
-72 copy_page_range
-73 do_anonymous_page
-76 get_empty_filp
-98 filemap_nopage
-168 unmap_all_pages
-352 find_get_page
-402 kmem_cache_free
-594 __pte_chain_free
-4360 page_add_rmap
-13542 page_remove_rmap
-17060 total

I'll try out akpm's file table stuff later today.

M.