From: Andrea Arcangeli
Date: Thu Mar 25 2004 - 18:41:16 EST

Ok, this seems feature complete. Both nonlinear swapping and prio_tree
are available now. I believe objrmap-core+anon-vma+prio_tree can be
merged into mainline after a bit more of testing, certainly they looks
good enough for -mm.

As usual this work wouldn't been possible without the efforts of Hugh
Dickins, Dave McCracken, and last but not the least Rajesh

As usual I'm writing this on my desktop while running this kernel and it
swaps fine (uptime still 10 minutes though ;).

after some misc swapping (desktop like load) this is the profiling:

724627 total 0.2810
613729 default_idle 9589.5156
47565 mmx_clear_page 424.6875
9710 do_page_fault 7.2898
5965 buffered_rmqueue 9.8109
4589 delay_tsc 143.4062
3986 do_no_page 2.1853
2679 unmap_page_range 3.4171
2487 free_hot_cold_page 9.1434
2363 page_add_rmap 6.4212
1897 page_address 10.7784
1600 handle_mm_fault 0.6993
1594 release_pages 4.5284
1289 page_remove_rmap 5.7545
1229 lru_cache_add_active 15.3625
1041 __alloc_pages 1.3555
1002 find_pte 5.2188
952 __copy_to_user_ll 3.9667
858 __pagevec_lru_add_active 4.4688
720 probe_irq_on 2.2500

the prio_tree funcs are at the end, this is expected since I have normal

14 prio_tree_insert 0.0203
8 prio_tree_first 0.0185
6 prio_tree_remove 0.0179
4 __vma_prio_tree_insert 0.0139
4 prio_tree_next 0.0081
2 __vma_prio_tree_remove 0.0054

the db simulator is running fine on the test box too.



Changelog diff between 2.6.5-rc2-aa2 and 2.6.5-rc2-aa3:

Only in 2.6.5-rc2-aa2: anon_vma.gz
Only in 2.6.5-rc2-aa3: anon-vma.gz

Rediffed to fixup a reject after objrmap-core update.

Files 2.6.5-rc2-aa2/extraversion and 2.6.5-rc2-aa3/extraversion differ
Files 2.6.5-rc2-aa2/objrmap-core.gz and 2.6.5-rc2-aa3/objrmap-core.gz differ

Fixup locking in swapoff, and microscalability optimization
in do_munmap. Both from the original objrmap patch from Dave McCracken.

Only in 2.6.5-rc2-aa3: prio-tree.gz

Drop computational complexity of swapping using objrmap, using
a prio tree for efficient search of the vma-ranges mapping
any given page in an inode. From Rajesh Venkatasubramanian.

More details can be found at his URL:
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/