[ 
https://issues.apache.org/jira/browse/HDFS-13702?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16524147#comment-16524147
 ] 

Todd Lipcon commented on HDFS-13702:
------------------------------------

Here are some perf results based on a 180GB teravalidate on a small cluster, as 
well as a 1GB teravalidate on LocalJobRunner (against a remote HDFS)
{code}
3.0.x original
------------------------------------
Avg map: 18sec
CPU time spent (ms)    2,208,950
GC time elapsed (ms)    68,153
 Performance counter stats for './run-validate.sh' (5 runs):

      22357.081985      task-clock (msec)         #    2.688 CPUs utilized      
      ( +-  6.78% )
            21,573      context-switches          #    0.965 K/sec              
      ( +-  2.58% )
             1,300      cpu-migrations            #    0.058 K/sec              
      ( +-  4.82% )
           425,146      page-faults               #    0.019 M/sec              
      ( +-  4.52% )
    63,809,409,850      cycles                    #    2.854 GHz                
      ( +-  6.56% )
    66,580,182,677      instructions              #    1.04  insn per cycle     
      ( +-  2.28% )
    13,489,574,848      branches                  #  603.369 M/sec              
      ( +-  4.58% )
       158,670,595      branch-misses             #    1.18% of all branches    
      ( +-  0.35% )

       8.317048233 seconds time elapsed                                         
 ( +-  0.10% )


3.0.x patched:
--------------------------------------------------------------------------------
Avg map time: 14sec
CPU time spent (ms)       1,750,180
GC time elapsed (ms)     42,468

 Performance counter stats for './run-validate.sh' (5 runs):

      14466.559412      task-clock (msec)         #    2.006 CPUs utilized      
      ( +-  3.18% )
            21,666      context-switches          #    0.001 M/sec              
      ( +-  0.55% )
             1,180      cpu-migrations            #    0.082 K/sec              
      ( +-  1.91% )
           234,159      page-faults               #    0.016 M/sec              
      ( +-  0.60% )
    41,793,452,250      cycles                    #    2.889 GHz                
      ( +-  2.77% )
    55,219,815,925      instructions              #    1.32  insn per cycle     
      ( +-  1.67% )
     9,837,238,534      branches                  #  679.998 M/sec              
      ( +-  2.57% )
       161,071,903      branch-misses             #    1.64% of all branches    
      ( +-  0.62% )

       7.210730451 seconds time elapsed                                         
 ( +-  0.25% )
{code}

> HTrace hooks taking 10-15% CPU in DFS client when disabled
> ----------------------------------------------------------
>
>                 Key: HDFS-13702
>                 URL: https://issues.apache.org/jira/browse/HDFS-13702
>             Project: Hadoop HDFS
>          Issue Type: Bug
>          Components: performance
>    Affects Versions: 3.0.0
>            Reporter: Todd Lipcon
>            Assignee: Todd Lipcon
>            Priority: Major
>         Attachments: hdfs-13702.patch
>
>
> I am seeing DFSClient.newReaderTraceScope take ~15% CPU in a teravalidate 
> workload even when HTrace is disabled. This is because it stringifies several 
> integers. We should avoid all allocation and stringification when htrace is 
> disabled.



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

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

Reply via email to