wsry commented on issue #11155: [FLINK-14818][benchmark] Fix receiving 
InputGate setup of StreamNetworkBenchmarkEnvironment.
URL: https://github.com/apache/flink/pull/11155#issuecomment-590666593
 
 
   I find that the number of tcp connections is one of the factors which 
influence the performance. If I change the connectionIndex from gateIndex to 
channelIndex which means it remains unchanged compared with the behavior before 
this change (setup 1000 tcp connections per gate if the number of channels is 
1000), the performance results are as follows. From the results, we can see 
that the performance goes down a bit compared with the results before this 
change, which should be caused by the decrease of floating buffers. Then what 
do you think? Should we keep the number of connections unchanged in this PR? 
@zhijiangW @pnowojski 
   ```
   Benchmark                                                                    
 (channelsFlushTimeout)   Mode  Cnt      Score      Error   Units
   DataSkewStreamNetworkThroughputBenchmarkExecutor.networkSkewedThroughput     
                    N/A  thrpt   30  17345.574 ±  370.647  ops/ms
   StreamNetworkBroadcastThroughputBenchmarkExecutor.networkBroadcastThroughput 
                    N/A  thrpt   30    608.881 ±   12.054  ops/ms
   StreamNetworkThroughputBenchmarkExecutor.networkThroughput                   
              100,100ms  thrpt   30  41732.518 ± 1109.436  ops/ms
   StreamNetworkThroughputBenchmarkExecutor.networkThroughput                   
          100,100ms,SSL  thrpt   30   9689.525 ±  202.895  ops/ms
   StreamNetworkThroughputBenchmarkExecutor.networkThroughput                   
               1000,1ms  thrpt   30  24106.705 ± 2952.364  ops/ms
   StreamNetworkThroughputBenchmarkExecutor.networkThroughput                   
             1000,100ms  thrpt   30  27509.665 ± 3246.965  ops/ms
   StreamNetworkThroughputBenchmarkExecutor.networkThroughput                   
         1000,100ms,SSL  thrpt   30   7691.287 ±  927.775  ops/ms
   StreamNetworkLatencyBenchmarkExecutor.networkLatency1to1                     
                    N/A   avgt   30     12.758 ±    0.147   ms/op
   ```

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


With regards,
Apache Git Services

Reply via email to