Versions Compared

Key

  • This line was added.
  • This line was removed.
  • Formatting was changed.

...

View file
nameDCAEGEN2 VES performance tests.odp
height250


Replacing Cambria with DMaaP Client

Presentation


View file
nameVesCollectorEventBatchAndNewClient.pptx
height250


Performance Tests with real DMaaP

Environment 

  • CPU - 24 cores
  • CPU clock speed - 2.4 GHz
  • Max Heap - 512 MB
  • Start Heap - 256 MB

Ves with Dmaap client

3_test_scenario_50_rps_time_600.jmx50RPS → 10min

Image Added

Image Added

Image Added

3_test_scenario_50_rps_time_600.jmx50RPS → 10min

Image Added

Image Added

Image Added

3_test_scenario_50_rps_time_600.jmx50RPS → 10min

Image Added

Image Added

Image Added

3_test_scenario_100_rps_time_600.jmx100RPS → 10min

Image Added

Image Added

Image Added

3_test_scenario_100_rps_time_600.jmx100RPS → 10min

Image Added

Image Added

Image Added

3_test_scenario_100_rps_time_600.jmx100RPS → 10min

Image Added

Image Added

Image Added


Ves with Cambria client

Ves version :  1.9.1


3_test_scenario_50_rps_time_600.jmx50RPS → 10min

Image Added

Image Added

Image Added

3_test_scenario_50_rps_time_600.jmx50RPS → 10min

Image Added

Image Added

Image Added

3_test_scenario_50_rps_time_600.jmx50RPS → 10min

Image Added

Image Added

Image Added

3_test_scenario_100_rps_time_600.jmx100RPS → 10min

Image Added

Image Added

Image Added

3_test_scenario_100_rps_time_600.jmx100RPS → 10min

Image Added

Image Added

Image Added

3_test_scenario_100_rps_time_600.jmx100RPS → 10min

Image Added

Image Added

Image Added


Summary test results:

Environment 

  • CPU - 24 cores
  • CPU clock speed - 2.4 GHz
  • Max Heap - 512 MB
  • Start Heap - 256 MB




Average and Max Sync Processing Time (Client → VES)Average VES Processing timeError Rate [%]Max CPU Usage [%]
VES with Dmaap50th percentile95th percentile99th percentile


50RPS → 10min122ms, 349ms181ms, 3.23s591ms, 4.43s91ms036
124ms, 515ms204ms, 4.27s609ms, 5.48s90ms042
121ms, 399ms177ms, 2.17s561ms, 5.2291ms030
100RPS → 10min274ms, 7.06570ms, 7.93s1.0s, 8.06s139ms0.05%72
574ms, 6,07s1,17s, 14.19s1.89s, 15.37s201ms091
291ms, 5.88s415.5ms, 6.45s922ms, 11.3s143ms078
VES with Cambria50th percentile95th percentile99th percentile


50RPS → 10min118ms, 520ms174ms, 2.10s571ms, 6.48s90ms048
122ms, 548ms230ms, 5.17s581ms, 5.54s88ms032
123ms, 557ms194ms, 2.50s676ms, 5.12s88ms042
100RPS → 10min301ms, 5.79s772ms, 16.97s1.16s, 17.08s153ms088
340ms, 7.13s636ms, 17.87s1.15s, 18.48s149ms076
307ms, 8.29s506ms, 9.44s855ms, 9.78s155ms088



Conclusion:

Results of performance tests for both VES collector client implementation (DMaap client and Cambria client) are very similar.

Max CPU usage, Error rate, Average VES processing time, average and max sync processing time(Client → Ves) are almost the same.


In Ves collector with DMaap client(100RPS for 10 min),  appeared an error with the connection pool. In that specific case, we had a connection poll set to 16 and we got an error that the connection poll limits have been reached (stack trace in attachment).

We have to handle that kind of error in code.