This MR aims at solving the following issue. Our experiments on active deep learning require to launch small simulations like Lorenz's attractor. These simulations are so small that there is no need for HPC, which is the core target of Melissa. We would like to run those simulations locally. To run many simulations locally can overload computer resources and result in the failure of the experiment. We should thus ensure the simulations are run given a computational budget. This would also be the occasion to provide a test case of Melissa for any user who does not have access to cluster.
The MR allows to run a fixed number of clients in parallel. It should not impact the behaviour of Melissa on real cluster environment.