Optimal information dispersal for probabilistic latency targets

The goal of this work is to provide a framework for applications to probabilistically meet latency targets using redundancy and multipath routing. It investigates the amount of redundancy necessary to maximize the probability of constructing a copy of a file F no later than a given delay bound Δ without any reservations. Redundancy is obtained from the information dispersal algorithm (IDA) of Rabin. Specifically, F is broken into b pieces, each of size s, and each piece is sent on one of the several (disjoint) paths from source to destination. Each piece contains some redundant information in such a way that any m ≡ |F|/s pieces are enough to reconstruct the file. This work presents both analytical and computational results to determine the optimal values of b and s to maximize the success probability. It also examines the impact of various parameters such as the file size, delay bound, loss rate, distance from the destination, on the probabilistic quality of service (QoS) guarantee. This is a novel approach to QoS routing since it does not require any reservations or extra overhead at intermediate nodes and effectively shifts the burden from nodes to links.