Beyond Wires: The Future of Interconnects

he first time I paid attention to interconnects in computing systems was during my graduate studies in the early 1990s at the University of Michigan, where I conducted research in parallel computing under the supervision of Professor Ed Davidson.In Ed's research group, most of us focused on performance analysis methodology (which Ed called "a functional approach") and parallelization techniques to understand and to accelerate large-scale simulation applications.The two projects I worked on included a car crash simulator sponsored by a nearby automobile company and an Indian Ocean circulation simulator, both were targeted to run on the Kendall Square Research machines and IBM SP2, a distributed memory machine using message-passing for shared data communication.Performance overheads due to node-tonode interprocessor communication, particularly in the IBM SP2, were always a top priority left to application programmers to manage, hide, and mitigate.