StaccatoLab: A Programming and Execution Model for Large-scale Dataflow Computing

Large-scale parallel computing is about organizing millions of arithmetic operations in parallel, each clock cycle. The dataflow programming model is then reviewed as a candidate for large-scale computing. StaccatoLab is introduced as a highly expressive dataflow programming language, supporting synchronous dataflow, cyclo-static dataflow, Boolean dataflow and non-determinate dataflow. The StaccatoLab execution model supports fine-grained (clock-cycle level) synchronization between processes, and minimizes memory usage. The execution model is described in terms of the basic dataflow theory and in part by augmenting the dataflow graph. Most model features extend to the more general forms of static and dynamic dataflows in a fairly straightforward fashion. After a description of the StaccatoLab tools, including Verilog synthesis, the chapter ends with a discussion on the outlook for dataflow programming for large-scale parallel computing.