Enhancing supercomputing with software defined networking

Supercomputers have long been characterized by their closed and proprietary interconnects. Software Defined Networking (SDN) allows for the programmatic flow management of network traffic. The introduction of SDN into the High Performance Computing (HPC) world permits scientists to leverage a high speed and simplified model for access to the immense computing power of supercomputers. As increasingly larger experimental and observational sites come online, scientists have a need to not only store data, but to compute upon that data in near real-time. SDN provides an avenue for the data from these experiments to flow directly into the supercomputer, without having to be staged first at an intermediary site, thus reducing the time to computation for that data. The newly gleaned results can then be used to guide the experiments in near real-time, allowing for more accurate data. This paper expands this vision, describing some of the motivating use cases in more detail, delineating our proposed architecture and implementation, chronicling our progress to date, and outlining future plans.