Parallel and Configurable Protocols: Experiences with a Prototype and an Architectural Framework

Abstract

As network speeds increase, a major concern regarding communication protocols is their limited throughput and latency due to processing overheads at network nodes. Furthermore, novel network applications and expanded network usage are leading to increased network requirements, particularly with respect to security and bandwidth. In this paper we consider the use of parallelism and configurability to increase throughput and reduce protocol processing latencies. We obtain experimental results of parallel protocol performance using a prototype implemented on a shared memory multiprocessor. The results demonstrate the utility of parallel protocol processing, and they indicate the further research necessary for constructing viable communication protocols for large-scale parallel machines. Based on these experiences, we present the design of an object-oriented framework for parallel protocol programming. This framework both facilitates parallel protocol development and helps maximize protocol performance on a wide variety of multiprocessors.