Balancing SRPT prioritization vs opportunistic gain in wireless systems with flow dynamics

The problem of scheduling best effort flows from a dynamic population sharing a time varying wireless channel, with the objective of minimizing mean sojourn time is considered. The key tradeoff involved is between prioritizing flows with short residual sizes and maximizing opportunistic capacity gain by selecting flows that currently see good channels. This tradeoff is explicitly characterized by introducing a new queueing model that involves servers with state-dependent capacity regions. In the transient case and for (bounding) polymatroid capacity regions, the optimal scheduler is given and used to obtain sub-optimality bounds for various heuristics. Using a mix of analysis and simulation two regimes are described: one where fully exploiting opportunistic gain is sufficient and further using residual flow sizes for scheduling will result in only minimal reduction in mean sojourn time, and the other, where the use of this information can indeed offer significant reduction. A new scheduler is proposed which performs well in both regimes.