You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@jmeter.apache.org by Philippe Mouawad <p....@ubik-ingenierie.com> on 2019/06/08 13:23:56 UTC

Improvement ideas for Distributed mode

Hello,

Nowadays, distributed testing is frequently done in the cloud or using
Docker or Kubernetes/Rancher OS.

I see the following fields of improvements , just throwing some ideas that
need further thinking:

   - Have a new mode that does not send full results to controller but only
   statistical data, keeping results locally as CSV. It would be similar to
   StatisticalSampleSender but overcoming its current issues and limitations:
      - Benefits:
         - Reduce drastically networks traffic between nodes and master to
         only few statistical data making it much more scalable
         - Questions/ Issues:
      - How would we compute Percentiles ? can we ?
      - Making Controller unaware of Worker nodes (jmeter-server) and
   making worker nodes register to it instead. Workflow would be:
      1. Start controller
         2. Start X Nodes
         3. Start test
         4. Start Y nodes implies adding dynamically more load
         - Benefits:
         - Only nodes are aware of master which simplifies configuration
         and communication
         - Allow dynamic addition of nodes on start
      - Change what worker nodes send to Controller:
      - Make them send only metrics and no more serializable data
      - Rely on Micrometer for that



Regards
Philippe M.




<https://www.openstreetmap.org/#map=18/50.69454/3.16455>