| A variety of real-time protocols have been developed to deal with communication failures in networked computing systems. These protocols rely on temporal and/or spatial redundancy to accomplish their goals--often at an expensive cost, due to inefficient usage of resources. Protocols that employ spatial redundancy, such as techniques based on resource-reservation and information redundancy (e.g., FEC), improve response time, but suffer from low resource utilization, with possible degradation in reliability. Alternately, protocols that employ temporal redundancy provide a high level of resource utilization and reliability, but tend to suffer from increased message delays and delay variances (i.e., jitter).; In this dissertation, we introduce AFTER--a framework for Adaptive Forward Timely Erasure Recovery (AFTER) that uses dynamic redundancy control to enhance both response time and jitter, while maintaining a high level of reliability and resource utilization. The main idea behind AFTER is to provide a flexible resource control mechanism for clients that require different levels of reliability and responsiveness, by dynamically balancing the levels of spatial and temporal redundancy used to deal with communication failures. In addition, AFTER aims to reduce the wasted bandwidth found in conventional FEC schemes, and at the same time, to enhance the timeliness and data integrity in message delivery. AFTER's dynamics redundancy control scheme is an elaboration of the adaptive redundancy scheme introduced in the Adaptive Information Dispersal Algorithm (AIDA). AFTER can be implemented in unreliable message transmissions as well as reliable transmissions, under various network environments ranging from high-speed B-ISDN networks to low-cost, best-effort communication channels.; The flexibility and superiority of AFTER is demonstrated through an implementation, TCP-Boston--a TCP/IP protocol for datagram transmission over ATM's ABR service. TCP-Boston shows AFTER's adaptability to best-effort traffic using reliable transport method under unreliable communication channels. Along with the protocol details of AFTER and implementation outlines of TCP-Boston, we present detailed simulation and analytic results that show the superiority of the AFTER-based protocol when compared to other methods, focusing on the real-time performance metrics, including message delay, delay-variance, percent of missed deadlines, as well as effective throughput. |
