Measurement of parallelism in data communications protocol software

The authors report on an experimental investigation of a parallel implementation of OSI protocol software which, unlike a single application, is a continuous flow system in which the program runs forever. An important attribute of parallel jobs, the average parallelism measure is extended to the domain of continuous flow systems, using the notion of the total rate of work. Protocol execution is important because it is known to have a significant impact on the overall performance of distributed systems. Experiments were performed on a parallel protocol implementation to measure the impact of the average and distribution of message size, the transport protocol data unit (TPDU) size, the flow control window size and the number of processors on parallelism and speedup. They indicate that parallelism and speedup depend more strongly on the design of the software and on its operating environment than on the protocol parameters.<
>