[ 
https://issues.apache.org/jira/browse/HADOOP-15549?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
 ]

Todd Lipcon updated HADOOP-15549:
---------------------------------
    Status: Patch Available  (was: Open)

Re-measured on trunk instead of 3.0.0, and also used 'taskset -c 1' to bind 
everything to one core to try to reduce variance:

2.8.2:
{code}
        535.052533      task-clock (msec)         #    0.918 CPUs utilized      
      ( +-  0.88% )
             3,055      context-switches          #    0.006 M/sec              
      ( +-  1.02% )
                 1      cpu-migrations            #    0.002 K/sec              
    
            12,644      page-faults               #    0.024 M/sec              
      ( +-  0.06% )
     1,953,309,627      cycles                    #    3.651 GHz                
      ( +-  0.16% )
     2,221,327,797      instructions              #    1.14  insn per cycle     
      ( +-  0.17% )
       417,919,978      branches                  #  781.082 M/sec              
      ( +-  0.17% )
        18,726,810      branch-misses             #    4.48% of all branches    
      ( +-  0.19% )

       0.582855783 seconds time elapsed                                         
 ( +-  1.70% )
{code}

3.2 without patch:
{code}
        751.038338      task-clock (msec)         #    0.927 CPUs utilized      
      ( +-  0.43% )
             3,646      context-switches          #    0.005 M/sec              
      ( +-  2.26% )
                 1      cpu-migrations            #    0.001 K/sec              
      ( +- 25.00% )
            19,233      page-faults               #    0.026 M/sec              
      ( +-  0.42% )
     2,735,218,817      cycles                    #    3.642 GHz                
      ( +-  0.62% )
     3,218,012,767      instructions              #    1.18  insn per cycle     
      ( +-  0.54% )
       604,477,739      branches                  #  804.856 M/sec              
      ( +-  0.48% )
        25,664,033      branch-misses             #    4.25% of all branches    
      ( +-  0.55% )

       0.810230298 seconds time elapsed                                         
 ( +-  0.98% )
{code}

3.2 with patch:
{code}
3.2 with patch:


        679.940626      task-clock (msec)         #    0.918 CPUs utilized      
      ( +-  1.14% )
             3,302      context-switches          #    0.005 M/sec              
      ( +-  1.16% )
                 1      cpu-migrations            #    0.001 K/sec              
    
            16,819      page-faults               #    0.025 M/sec              
      ( +-  0.06% )
     2,375,283,537      cycles                    #    3.493 GHz                
      ( +-  0.33% )
     2,722,724,476      instructions              #    1.15  insn per cycle     
      ( +-  0.27% )
       511,944,028      branches                  #  752.925 M/sec              
      ( +-  0.24% )
        21,981,131      branch-misses             #    4.29% of all branches    
      ( +-  0.33% )

       0.740316578 seconds time elapsed                                         
 ( +-  0.96% )
{code}

> Upgrade to commons-configuration 2.1 regresses task CPU consumption
> -------------------------------------------------------------------
>
>                 Key: HADOOP-15549
>                 URL: https://issues.apache.org/jira/browse/HADOOP-15549
>             Project: Hadoop Common
>          Issue Type: Bug
>          Components: metrics
>    Affects Versions: 3.0.2
>            Reporter: Todd Lipcon
>            Assignee: Todd Lipcon
>            Priority: Major
>         Attachments: hadoop-15549.txt
>
>
> HADOOP-13660 upgraded from commons-configuration 1.x to 2.x. 
> commons-configuration is used when parsing the metrics configuration 
> properties file. The new builder API used in the new version apparently makes 
> use of a bunch of very bloated reflection and classloading nonsense to 
> achieve the same goal, and this results in a regression of >100ms of CPU time 
> as measured by a program which simply initializes DefaultMetricsSystem.
> This isn't a big deal for long-running daemons, but for MR tasks which might 
> only run a few seconds on poorly-tuned jobs, this can be noticeable.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)

---------------------------------------------------------------------
To unsubscribe, e-mail: common-issues-unsubscr...@hadoop.apache.org
For additional commands, e-mail: common-issues-h...@hadoop.apache.org

Reply via email to