Dynamic Server Allocation over Time Varying Channels with Switchover Delay

TL;DR

This paper studies dynamic server allocation in systems with time-varying channels and switchover delays, introducing novel policies that optimize throughput and stability in such complex environments.

Contribution

It introduces a new framework for analyzing stability and designing throughput-optimal policies considering switchover delays and varying connectivity.

Findings

FBDC policy is asymptotically throughput-optimal.

Myopic policies achieve over 90% of the stability region.

Explicit stability region characterization for two-queue systems.

Abstract

We consider a dynamic server allocation problem over parallel queues with randomly varying connectivity and server switchover delay between the queues. At each time slot the server decides either to stay with the current queue or switch to another queue based on the current connectivity and the queue length information. Switchover delay occurs in many telecommunications applications and is a new modeling component of this problem that has not been previously addressed. We show that the simultaneous presence of randomly varying connectivity and switchover delay changes the system stability region and the structure of optimal policies. In the first part of the paper, we consider a system of two parallel queues, and develop a novel approach to explicitly characterize the stability region of the system using state-action frequencies which are stationary solutions to a Markov Decision Process (MDP) formulation. We then develop a frame-based dynamic control (FBDC) policy, based on the state-action frequencies, and show that it is throughput-optimal asymptotically in the frame length. The FBDC policy is applicable to a broad class of network control systems and provides a new framework for developing throughput-optimal network control policies using state-action frequencies. Furthermore, we develop simple Myopic policies that provably achieve more than 90% of the stability region. In the second part of the paper, we extend our results to systems with an arbitrary but finite number of queues.