Computing

At hardware level, the computing part of MUST is currently provided by a set of 200 servers supplying around 100000 HEPSPEC06 (~140 TFlops) and allowing to run more than 7000 simultaneous computing tasks (jobs), each of them being able to use at least 2 GB of memory. Over 92 % of computing resources are constantly used, mainly because of the intensive LHC computing.

Each of these computing servers is connected through 10Gb/s in order to handle the flow generated by access to data, supporting both IPv4 and IPv6 traffic.

At software level, a RHEL9 compatible Linux operating system is installed on all MUST elements. Ressources are also accessible through grid protocols using UMD-5 middleware.

A HPC-compatible platform

All of the computing servers of the MUST platform allow HPC (High Performance Computing) via the execution of parallel computations based on the MPI communication standard.

A platform including GPU cards

Nine computing servers include NVIDIA Tesla K80, Tesla V100, Quadro P6000, Tesla T4 and Ampere A100 GPU cards. They allow, among other things, to perform vector-type calculations, deep learning and image processing.