Scheduling algorithms for throughput maximization in time-varying networks with reconfiguration delays

Abstract

We consider the control of possibly time-varying wireless networks under reconfiguration delays. Reconfiguration delay is the time it takes to switch network resources from one subset of nodes to another and it is a widespread phenomenon observed in many practical systems. Optimal control of networks has been studied to a great extent in the literature, however, the significant effects of reconfiguration delays received limited attention. Moreover, simultaneous presence of time-varying channels and reconfiguration delays has never been considered and we show that it impacts the system fundamentally. We first consider a Delay Tolerant Network model where data messages arriving randomly in time and space are collected by mobile collectors. In this setting reconfiguration delays correspond to travel times of collectors. We utilize a combination of wireless transmission and controlled mobility to improve the system delay scaling with load [rho] from [theta](1/(1-[rho])²) to [theta](1/1-[rho]), where the former is the delay for the corresponding system without wireless transmission. We propose control algorithms that stabilize the system whenever possible and have optimal delay scaling. Next, we consider a general queuing network model under reconfiguration delays and interference constraints which includes wireless, satellite and optical networks as special cases. We characterize the impacts of reconfiguration delays on system stability and delay, and propose scheduling algorithms that persist with service schedules for durations of time based on queue lengths to minimize negative impacts of reconfiguration delays. These algorithms provide throughput-optimality without requiring knowledge of arrival rates since they dynamically adapt inter-switching durations to stochastic arrivals. Finally, we present optimal scheduling under time-varying channels and reconfiguration delays, which is the main contribution of this thesis. We show that under the simultaneous presence of these two phenomenon network stability region shrinks, previously suggested policies are unstable, and new algorithmic approaches are necessary. We propose techniques based on state-action frequencies of Markov Decision Process theory to characterize the network stability region and propose throughput-optimal algorithms. The state-action frequency technique is applicable to a broad class of systems with or without reconfiguration delays, and provides a new framework for characterizing network stability region and developing throughput-optimal scheduling policies.