The Evolution of a Distributed Dataflow Processing Model using Ada Scott James Management Communications and Control, Inc. Suite 220 2000 N. 14th Street Arlington VA 22201 james@mcci-arl-va.com Abstract This paper presents the stages of design for a dataf3ow program. A sequence of models is presented in increasing order of complexity, demonstrating the values and shortcomings of each. Techniques and challenges mapping the dataflow paradigm to the Ada concurrency model and distribution annex are described. Nodes with no upstream queues are said to be sources and generate data autonomously. Nodes with no downstream queues are said to be sinks and can be thought to discard or display the data. Completely disconnected nodes are of no interest to the model. Our goal is to model this behavior, providing a system with no deadlocks and minimal bottlenecks to dataflow. For purposes of analysis, the node processing time will be loosely considered significant compared to the data transfer time, thus we wish to minimize any delays on data transfer due to node processing. Definitions and Intent A dataflov graph [4] can be thought of as plumbing. Information flows as water through the piping (queues) is stored and released from various sinks, fixtures,
/lp/association-for-computing-machinery/the-evolution-of-a-distributed-dataflow-processing-model-using-ada-5sOwJ6RjHf