Asynchrony in Parallel Computing: From Dataflow to Multithreading

The paper presents an overview of the parallel computingmodels, architectures, and research projects that are based on asynchronous instruction scheduling. It starts with pure data ow computing models and presents an historical development of several ideas (i.e. single-token-per-arc data ow, tagged-token data ow, explicit token store, threaded data ow, large-grain data ow, RISC data ow, cycle-by-cycle interleaved multithreading, block interleaved multithreading, simultaneous multithreading) that resulted in modernmultithreaded superscalar processors. The paper shows that uni cation of von Neumann and data ow models is possible and preferred to treating them as two unrelated, orthogonal computing paradigms. Today's data ow research incorporates more explicit notions of state into the architecture, and von Neumann models using many data ow techniques to improve the latency hiding aspects of modern multithreaded systems.