Sampling Distributed Schedulers for Resilient Space Communication

Abstract

We consider routing in delay-tolerant networks like satellite constellations with known but intermittent contacts, random message loss, and resource-constrained nodes. Using a Markov decision process model, we seek a forwarding strategy that maximises the probability of delivering a message given a bound on the network-wide number of message copies. Standard probabilistic model checking would compute strategies that use global information, which are not implementable since nodes can only act on local data. In this paper, we propose notions of distributed schedulers and good-for-distributed-scheduling models to formally describe an implementable and practically desirable class of strategies. The schedulers consist of one sub-scheduler per node whose input is limited to local information; good models additionally render the ordering of independent steps irrelevant. We adapt the lightweight scheduler sampling technique in statistical model checking to work for distributed schedulers and evaluate the approach, implemented in the Modest Toolset, on a realistic satellite constellation and contact plan.