[GitHub] [kafka] hachikuji commented on pull request #9563: KAFKA-10684; Avoid additional envelope copies during network transmission

[GitBox]

hachikuji commented on pull request #9563:
URL: https://github.com/apache/kafka/pull/9563#issuecomment-723209437


   Here are updated benchmarks for `FetchResponseBenchmark`.
   
   Trunk:
   ```
   Benchmark                                          (partitionCount)  
(topicCount)  Mode  Cnt        Score       Error  Units
   FetchResponseBenchmark.testSerializeFetchResponse                 3          
  10  avgt   15     6037.058 ±    41.647  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                 3          
 500  avgt   15   296913.061 ±  2548.748  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                 3          
1000  avgt   15   607466.969 ±  5149.176  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                10          
  10  avgt   15    17971.239 ±   141.934  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                10          
 500  avgt   15   888464.188 ±  6341.169  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                10          
1000  avgt   15  1938108.325 ± 32056.839  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                20          
  10  avgt   15    34217.206 ±   234.166  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                20          
 500  avgt   15  1830593.478 ± 22601.172  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                20          
1000  avgt   15  4983809.392 ± 38217.505  ns/op
   ```
   
   This patch:
   ```
   Benchmark                                          (partitionCount)  
(topicCount)  Mode  Cnt        Score       Error  Units
   FetchResponseBenchmark.testSerializeFetchResponse                 3          
  10  avgt   15     5686.644 ±    92.892  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                 3          
 500  avgt   15   294136.644 ±  6709.506  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                 3          
1000  avgt   15   599678.709 ±  5837.803  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                10          
  10  avgt   15    17065.100 ±   241.715  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                10          
 500  avgt   15   907448.421 ± 14203.990  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                10          
1000  avgt   15  1906411.458 ± 26450.041  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                20          
  10  avgt   15    33147.044 ±  1181.493  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                20          
 500  avgt   15  1861946.674 ± 25145.868  ns/op
   FetchResponseBenchmark.testSerializeFetchResponse                20          
1000  avgt   15  4339003.253 ± 69576.240  ns/op
   ```
   
   So consistently better, but not groundbreaking. That is fine since the main 
improvement is simplifying the generation of `Send` objects for all request 
types.


----------------------------------------------------------------
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:
us...@infra.apache.org