romeoLAB: A High Performance Training Platform for HPC, GPU and DeepLearning
Résumé
In this pre-exascale era, we are observing a dramatic increase of the necessity of computer science courses dedicated to parallel programming on heterogeneous architectures. The full hybrid cluster Romeo has been used in that purpose since a long time in order to train master students and cluster users. The main issue for trainees is the cost of accessing and exploiting a production facility in a pedagogic context. The use of some specific techniques and software (SSH, workload manager, remote file system, ...) is mandatory without being part of courses prerequisites nor pedagogic objectives. The romeoLAB platform we developed at ROMEO HPC Center is an online interactive pedagogic platform for HPC and GPU technologies courses. Its main purpose is to simplify the process of resources usage in order to focus on the taught subjects. This paper presents the romeoLAB architecture as well as its motivations, usages and future improvements.