Grid'5000 experiment

Jump to: navigation, search

Experiment Management for Grid'5000 (Middleware)

Conducted by

Brice Videau


We studied the reproducibility of experiments in a grid environment. We found that even nodes supposed to be identical sometimes performed very differently from one another. This variability was mainly due to hardware weakness, hardware differences or misconfiguration of some nodes. This variability proved to induce serious bias in experiments.




  • Nodes involved: 100
  • Sites involved: 2
  • Minimum walltime: 8h
  • Batch mode: no
  • CPU bound: yes
  • Memory bound: no
  • Storage bound: yes
  • Network bound: yes
  • Interlink bound: yes

Tools used

No information


This study showed that to obtain meaningfull experiment results one has to carefully study each node in order to prove them correctly working before the experiment takes place. The impossibility to achieve such a task by hand implies that some tools need to be developed to automate the process.

Shared by: Brice Videau
Last update: 0000-00-00 00:00:00
Experiment #67