This project is a collaboration of scientists working in two HBP Subprojects:
- HPAC Platform (SP7)
- Neurorobotics Platform (SP10)
The frontend of the Neurorobotics Platform (NRP) enables the user to design virtual experiments with robots, to start the simulation from the Collaboratory and to visualise the running experiment. The neural simulation backend can either be neuromorphic systems (SP9) or NEST. This backend is currently deployed on OpenStack Virtual Machines (VM) managed by Puppet. The VMs used for Gazebo, the robot simulation, additionally require access to GPUs. The NEST simulations require a few cluster nodes with MPI that are managed by Slurm.
The currently used compute resources limit the maximum number of concurrent users that can run their virtual experiments in parallel. Therefore, the same services should be deployed on the HPAC Platform as well to reach a higher scalability. This requires access to a few nodes of at least one HPC system, some of which should contain GPUs for running Gazebo. Additionally, suitable deployment mechanisms are required, which at least in a first phase needs to be container-based.
Problems to be solved
The following problems require special attention and need to be solved in co-design with the users for implementing the use case successfully.
- Enabling container-based deployment of the simulations on an HPC system, including GPU support for Docker.