Date of Award

Summer 1993

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Computer Science

Committee Director

Kurt Maly

Committee Member

Mohammed Zubair

Committee Member

Ravi Mukkamala

Abstract

The research reported in this thesis investigates the use of parallelism at multiple levels to realize high-speed networks that offer advantages in throughput, cost, reliability, and flexibility over alternative approaches. This research specifically considers use of parallelism at two levels: the "upper" level and the "lower" level. At the upper level, N protocol processors perform functions included in the transport and network layers. At the lower level, M channels provide data and physical layer functions. The resulting system provides very high bandwidth to an application. A key concept of this research is the use of replicated channels to provide a single, high bandwidth channel to a single application. The parallelism provided by the network is transparent to communicating applications, thus differentiating this strategy from schemes that provide a collection of disjoint channels between applications on different nodes. Another innovative aspect of this research is that parallelism is exploited at multiple layers of the network to provide high throughput not only at the physical layer, but also at upper protocol layers. Schedulers are used to distribute data from a single stream to multiple channels and to merge data from multiple channels to reconstruct a single coherent stream. High throughput is possible by providing the combined bandwidth of multiple channels to a single source and destination through use of parallelism at multiple protocol layers. This strategy is cost effective since systems can be built using standard technologies that benefit from the economies of a broad applications base. The exotic and revolutionary components needed in non-parallel approaches to build high speed networks are not required. The replicated channels can be used to achieve high reliability as well. Multilevel parallelism is flexible since the degree of parallelism provided at any level can be matched to protocol processing demands and application requirements.

DOI

10.25777/4eky-1111

Share

COinS