Github user roshannaik commented on the pull request:

    https://github.com/apache/storm/pull/1217#issuecomment-200543287
  
    @arunmahadevan  :-)  ...  i am not taking the throughput measurements while 
profiler is attached.
    It will take some time for me to continue to iterate over and analyze your 
attempts and JProfiler usage to see what is going on. With a quick glance I see 
multiple differences in your topology setup.  But the profiler screenshots that 
i have posted are hopefully evidence that I didn't cook it up :-). You can 
either try with the topology i described ..  also i shall post a Github link to 
the topology i am using soon.
    
    @HeartSaVioR I am a bit puzzled to see a 8% or  25% diff  in perf (for a 
given topology) being referred to as *micro* optimization.  This is a case of 
potentially significant overhead being imposed upon the common code path by a 
infrequently used code path.  Quite the contrary, i feel, one should have to 
have a very good justification to leave this turned on.  
    
    It is not feasible to do a full fledged Yahoo style benchmark to identify 
and fix all such issues. Micro-benchmarking is essential. Here we are looking 
at a simple case of emit() call dominating most of the time within nextTuple() 
...   the spout computation itself is taking negligible % of the time. 
    
    I have deliberately separated out #1242 from this .. as this is PR about 
simply disabling a DEBUG config setting.. as opposed to modifying code to avoid 
repetitive lookups. Seeking and testing an alternative implementation for event 
logging (unless its trivial) i felt might be tricky at this late stage of 1.x.
    
    



---
If your project is set up for it, you can reply to this email and have your
reply appear on GitHub as well. If your project does not have this feature
enabled and wishes so, or if the feature is enabled but not working, please
contact infrastructure at infrastruct...@apache.org or file a JIRA ticket
with INFRA.
---

Reply via email to