On the Throughput-Latency Routing Optimality of a Delay-Tolerant Network

In addition to the disconnected and dynamically changing graph that characterizes a wireless delay-tolerant network (DTN), the constrained storage capacity of the nodes can lead to buffer overflows. DTN routing is commonly designed to either minimize the end-to-end data forwarding latency or maximize throughput without paying much attention to the close interrelation between the two metrics. When used in isolation, results may be conflicting, e.g., achieve low latency by routing bundles through paths with a high drop rate, that is, low throughput. A comparative analysis of the theoretical performance of a latency-only, inverse throughput-only, and combined routing objective is provided by modeling the system with a finite Continuous Time Markov Chain. The Pareto front of the multi-objective latency-throughput routing case is discussed and approached with the inverse of Kleinrock's power metric and a simplified variant. Moreover, the conditions for a latency-throughput paradox in a DTN are discussed where the regular latency trend with higher traffic loads is reversed. The study brings new insight into the multi-objective DTN routing problem and implications in the design of robust path selection methods.