Enabling a SuperFacility with Software Defined Networking

Experimental and Observational facilities are increasingly turning to high-performance computing centers to meet their growing analysis requirements. The combination of experimental facilities with HPC Centers has been termed the Super Facility. This vision requires a new level of connectivity and bandwidth between these remote instruments and the HPC systems. NERSC in collaboration with Cray has been exploring a new model of networking that builds on the principles of Software Defined Networking. We envision an architecture that allows the wide-area network to extend into the Cray system and enables external facilities to stream data directly to compute resources inside the system at over a 100 Gbs in the near-future and eventually reach beyond a 1 Tbs. In this paper will expand on this vision, describe some of the motivating use cases in more detail, layout our proposed architecture and implementation, describe our progress to date, and outline future plans. Keywords-SDN; HPC systems