lkml.org 
[lkml]   [2011]   [Jun]   [17]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: REGRESSION: Performance regressions from switching anon_vma->lock to mutex
From
Date
On Fri, 2011-06-17 at 11:39 -0700, Linus Torvalds wrote:

> Having gone over it a bit more, I actually think I prefer to just
> special-case the allocation instead.
>
> We already have to drop the anon_vma lock for the "out of memory"
> case, and a slight re-organization of clone_anon_vma() makes it easy
> to just first try a NOIO allocation with the lock still held, and then
> if that fails do the "drop lock, retry, and hard-fail" case.
>
> IOW, something like the attached (on top of the patches already posted
> except for your memory reclaim thing)
>

Linus,

I've applied this patch, plus the other two patches on batching anon_vma
clone and anon_vma unlink. This improved throughput further. I now see
average throughput at 140.2% vs 2.6.39-vanilla over 10 runs. The mutex
lock has also gone down to 3.7% of cpu in my profile. Certainly a great
deal of improvements.

To summarize,

Throughput
2.6.39(vanilla) 100.0%
2.6.39+ra-patch 166.7% (+66.7%)
3.0-rc2(vanilla) 68.0% (-32%)
3.0-rc2+linus (v1) 115.7% (+15.7%) (anon_vma clone v1)
3.0-rc2+linus+softirq 86.2% (-17.3%)
3.0-rc2+linus (v2) 104.9% (+4.9%) (anon_vma clone v2)
3.0-rc2+linus (v3) 140.3% (+40.3%) (anon_vma clone v2 + unlink + chain_alloc_tweak)


(Max-Min)/avg Standard Dev
2.6.39(vanilla) 3% 1.1%
2.6.39+ra-patch 3% 1.2%
3.0-rc2(vanilla) 20% 7.3%
3.0-rc2+linus 36% 12.2%
3.0-rc2+linus+softirq 40% 15.2%
3.0-rc2+linus (v2) 53% 14.8%
3.0-rc2+linus (v3) 27% 8.1%

Thanks.

Tim


------------Profile attached--------------

Profile from latest run 3.0-rc2+linus (v3):

- 5.44% exim [kernel.kallsyms] [k] _raw_spin_lock_irqsave
- _raw_spin_lock_irqsave
+ 87.81% cpupri_set
+ 5.67% release_pages
+ 1.71% pagevec_lru_move_fn
+ 1.31% try_to_wake_up
+ 0.85% get_page_from_freelist
+ 4.15% exim [kernel.kallsyms] [k] page_fault
- 3.76% exim [kernel.kallsyms] [k] __mutex_lock_common.clone.5
- __mutex_lock_common.clone.5
- 99.97% __mutex_lock_slowpath
- mutex_lock
+ 55.46% lock_anon_vma_root.clone.13
+ 41.94% anon_vma_lock.clone.10
+ 1.14% dup_mm
+ 1.02% unlink_file_vma
+ 2.44% exim [kernel.kallsyms] [k] unmap_vmas
+ 2.06% exim [kernel.kallsyms] [k] do_raw_spin_lock
+ 1.91% exim [kernel.kallsyms] [k] page_cache_get_speculative
+ 1.89% exim [kernel.kallsyms] [k] copy_page_c





\
 
 \ /
  Last update: 2011-06-17 22:23    [W:0.092 / U:0.404 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site