pnowojski edited a comment on pull request #32:
URL: https://github.com/apache/flink-benchmarks/pull/32#issuecomment-927976175


   > the new one is run with parallelism = 4 (old one was run with 1)
   
   Can you compare the results with parallelism = 1?
   
   > we measure checkpoints only when there is high backpressure, in the old we 
don't really know to what extent the pipeline is saturated
   
   The old benchmark is doing many checkpoints, and the configuration was 
supposed to be tweaked in such a way, that the job should back pressure very 
very quickly, within first or a couple of first checkpoints. So the average 
number of checkpoints/s from old benchmark should be comparable to yours one 
show invocation time.
   
   The fact that you don't see a difference between timeout=0 and timeout=1 
suggests that something might be wrong with your setup. Both old micro 
benchmark and our previous full cluster tests were in an agreement that using 
timeouts significantly impacts checkpoint time. So replicating the old values 
in your new benchmark could help us validate some things.


-- 
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.

To unsubscribe, e-mail: [email protected]

For queries about this service, please contact Infrastructure at:
[email protected]


Reply via email to