Thread (86 messages) 86 messages, 13 authors, 2011-06-18

Re: REGRESSION: Performance regressions from switching anon_vma->lock to mutex

From: Peter Zijlstra <peterz@infradead.org>
Date: 2011-06-15 11:43:06
Also in: lkml

On Wed, 2011-06-15 at 12:58 +0200, Peter Zijlstra wrote:
On Wed, 2011-06-15 at 12:36 +0200, Peter Zijlstra wrote:
quoted
On Tue, 2011-06-14 at 17:29 -0700, Tim Chen wrote:
quoted
MOSBENCH test suite.
Argh, I'm trying to get this thing to run, but its all snake poo..

/me takes up a heavy club and goes snake hunting, should make a pretty
hat or something.
Sweet, I've got meself a snake-skin hat!

The first thing that stood out when running it was:

31694 root      20   0 26660 1460 1212 S 17.5  0.0   0:01.97 exim 
    7 root      -2  19     0    0    0 S 12.7  0.0   0:06.14 rcuc0 
   24 root      -2  19     0    0    0 S 11.7  0.0   0:04.15 rcuc3 
   34 root      -2  19     0    0    0 S 11.7  0.0   0:04.10 rcuc5 
   39 root      -2  19     0    0    0 S 11.7  0.0   0:06.38 rcuc6 
   44 root      -2  19     0    0    0 S 11.7  0.0   0:04.53 rcuc7 
   49 root      -2  19     0    0    0 S 11.7  0.0   0:04.11 rcuc8 
   79 root      -2  19     0    0    0 S 11.7  0.0   0:03.91 rcuc14 
   89 root      -2  19     0    0    0 S 11.7  0.0   0:03.90 rcuc16 
  110 root      -2  19     0    0    0 S 11.7  0.0   0:03.90 rcuc20 
  120 root      -2  19     0    0    0 S 11.7  0.0   0:03.82 rcuc22 
   13 root      -2  19     0    0    0 S 10.7  0.0   0:04.37 rcuc1 
   19 root      -2  19     0    0    0 S 10.7  0.0   0:04.19 rcuc2 
   29 root      -2  19     0    0    0 S 10.7  0.0   0:04.12 rcuc4 
   54 root      -2  19     0    0    0 S 10.7  0.0   0:04.11 rcuc9 
   59 root      -2  19     0    0    0 S 10.7  0.0   0:04.40 rcuc10 
   64 root      -2  19     0    0    0 R 10.7  0.0   0:04.17 rcuc11 
   69 root      -2  19     0    0    0 R 10.7  0.0   0:04.23 rcuc12 
   84 root      -2  19     0    0    0 S 10.7  0.0   0:03.90 rcuc15 
   95 root      -2  19     0    0    0 S 10.7  0.0   0:03.99 rcuc17 
  100 root      -2  19     0    0    0 S 10.7  0.0   0:03.88 rcuc18 
  105 root      -2  19     0    0    0 S 10.7  0.0   0:04.14 rcuc19 
  125 root      -2  19     0    0    0 S 10.7  0.0   0:03.79 rcuc23 
   74 root      -2  19     0    0    0 S  9.7  0.0   0:04.33 rcuc13 
  115 root      -2  19     0    0    0 R  9.7  0.0   0:03.82 rcuc21 

Which is an impressive amount of RCU usage..
FWIW, Alex Shi's patch:

http://lkml.kernel.org/r/1308029185.15392.147.camel@sli10-conroe

Improves the situation to:

 3745 root      20   0 26664 1460 1212 S 18.5  0.0   0:01.28 exim 
   39 root      -2  19     0    0    0 S  4.9  0.0   0:02.83 rcuc6 
  105 root      -2  19     0    0    0 S  4.9  0.0   0:02.79 rcuc19 
    7 root      -2  19     0    0    0 S  3.9  0.0   0:02.70 rcuc0 
   13 root      -2  19     0    0    0 S  3.9  0.0   0:02.54 rcuc1 
   19 root      -2  19     0    0    0 S  3.9  0.0   0:02.76 rcuc2 
   24 root      -2  19     0    0    0 S  3.9  0.0   0:02.75 rcuc3 
...

And throughput increases like:

-tip            260.092 messages/sec/core
-tip+sirq-rcu   271.078 messages/sec/core


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help