The Researcher's Guide to HPC Networks
Networks are critical infrastructure in nearly every computing system, but they are particularly important in High Performance Computing (HPC) systems, where users tackle problems at massive scale. Unfortunately, many of the hardware and software technologies that comprise supercomputer and datacenter networks are not widely taught in classes, nor are they extensively covered by textbooks or other educational documents outside of their primary sources. The purpose of this document is to provide a starting point to learn about these technologies by lowering the barrier to understanding and providing an abundance of references for further research. Covered topics include all levels of communication and network programming APIs, network control plane technologies, physical interconnects, and link-layer devices which are then brought to life in discussion of the architecture of El Capitan, a leadership class supercomputer created by Lawrence Livermore National Laboratory.