Lightweight Protocols for Distributed Systems

A methodology and architecture have been developed that contrast sharply with common interpretations of the Open Systems Interconnection concept of Layering. The thesis is that while many functionalities are required to support the wide range of services in a Distributed System, no more than two concurrent layers of protocol are required below the application. The rest of the functions required may be specified and implemented as in-line code, or procedure calls. The proliferation of similar functions such as multiplexing and error recovery, through more than one or two layers in the OSI model, contradicts many basic rules of modularisation, and leads to inherent inefficiency and lack of robustness. Furthermore, it makes the implementation of Open Systems in hardware almost intractable. A key design issue is how to construct the automatic adaptation mechanisms that will allow a protocol to match the service requirements of the user. The arguments in this thesis are based on the experience of the design and implementation of three major classes of protocol for distributed systems: a transaction protocol (ESP), a reliable multicast protocol (MSP), and a Stream Protocol (TCP). A number of performance measurements of these protocols, including the latter, the DoD Transmission Control Protocol, were carried out to support the thesis. Finally, the impact of these protocols and this architecture on the switching fabric are investigated. We examine mechanisms for congestion control and fair sharing of the bandwidth in a huge internet connected by datagram routers and MAC Bridges, with a variety of link technologies ranging from terrestrial to satellite. The discussion of Interconnection techniques is included because it is the view of die author that understanding traffic distribution algorithms is vital to understanding end to end protocol performance