Throughput Optimal and Fast Near-Optimal Scheduling with Heterogeneously Delayed Network-State Information (Extended Version)

Abstract

We consider the problem of distributed scheduling in wireless networks where heterogeneously delayed information about queue lengths and channel states of all links are available at all the transmitters. In an earlier work (by Reddy et al. in Queueing Systems, 2012), a throughput optimal scheduling policy (which we refer to henceforth as the R policy) for this setting was proposed. We study the R policy, and examine its two drawbacks -- (i) its huge computational complexity, and (ii) its non-optimal average per-packet queueing delay. We show that the R policy unnecessarily constrains itself to work with information that is more delayed than that afforded by the system. We propose a new policy that fully exploits the commonly available information, thereby greatly improving upon the computational complexity and the delay performance of the R policy. We show that our policy is throughput optimal. Our main contribution in this work is the design of two fast and near-throughput-optimal policies for this setting, whose explicit throughput and runtime performances we characterize analytically. While the R policy takes a few milliseconds to several tens of seconds to compute the schedule once (for varying number of links in the network), the running times of the proposed near-throughput-optimal algorithms range from a few microseconds to only a few hundred microseconds, and are thus suitable for practical implementation in networks with heterogeneously delayed information.