Grid'5000 experiment

Jump to: navigation, search

Stressing kadeploy with karate/kastress (Middleware)

Conducted by

Julien Leduc


Kadeploy had some concurrent issues reported by Arnault Fontaine, he developed kastress/karate to reproduce those problems. This serie of experiments permitted to qualify kadeploy 2.1.5 and 2.1.6 and validate fixes to concurrency issues. Kadeploy was qualified on 64, 128 and 256 nodes with 1, 2, 4 and 8 concurrent synchronized deployments on the same cluster, with kadeploy version 2.1.2, 2.1.5 and 2.1.6. The tests conducted showed that the expermental conditions raised concurrency issues on kadeploy 2.1.2, solved in both kadeploy 2.1.5 and kadeploy 2.1.6.




  • Nodes involved: 500
  • Sites involved: 1
  • Minimum walltime: 2h
  • Batch mode: yes
  • Use kadeploy: yes
  • CPU bound: no
  • Memory bound: no
  • Storage bound: no
  • Network bound: no
  • Interlink bound: no

Tools used

kastress/karate, kadeploy


Validation of the qualification test of kadeploy: a qualification test was developped to reproduce the conditions where an issue occured in the middleware, and lead to a solution. This approach gave very good results in the specific distributed environment of Grid'5000, and with asynchronous tests (the test developer has finished his internship period at the time of the final validation).

Shared by: Julien Leduc
Last update: 2007-07-30 11:18:24
Experiment #391