A Queue-length-based Randomized Scheduler for Wireless Networks

Dynamic queue-length-based schedulers have been shown to maximize the throughput performance of data networks. However, for interference-limited networks, such as wireless and sensor networks, their implementation is computationally intractable. To resolve this, randomized policies with both centralized and distributed implementations have been studied in the literature. In this paper, we propose and analyze a new randomized policy, called the RANDWEIGHT Scheduling Policy, that reduces the complexity of the scheduling component while providing throughput guarantees for the system. For a bipartite network topology, we study the stability region of our policy and show that under sufficiently symmetric loads the policy is guaranteed to stabilize the network. The policy presented here is centralized, however it is amenable to distributed implementation as discussed in the paper. Moreover, rate control can be jointly implemented with this policy using recent methods proposed in the context of cross-layer network control.