Optimal Control for Wireless Software Defined Networks: Theory and Implementation

Abstract

Wireless Software Defined Network (SDN) has emerged as a new programmable network paradigm that facilitates flexibility in robust control and management. Toward the production-level network deployment at scale, there has been a surge of interest in distributed architectures of wireless SDN. Despite the inherent importance, optimal network control for either centralized or distributed wireless SDN has remained an open problem, where previous works either fail to account for wireless interference constraints, or are only sub-optimal in throughput due to quasi-static shortest path routing. Though throughput-optimal and well-established in the literature, the BackPressure (BP) algorithm is not compatible with wireless SDN architecture. In contrast, the recently developed Universal Max-Weight (UMW) policy also achieves throughputoptimality, yet permits algorithmic structure more congruent with SDN’s requirements. Unlike BP, UMW pre-computes a fixed route per-packet upon a packet arrival, which can be integrated with the flow installation phase of SDN, and uses novel easy-to-track virtual queues in place of physical queues (of backlogged packets), whose operations are not supported by SDN switches. In this thesis, we propose novel UMW-based optimal control frameworks for both centralized and distributed wireless SDN that achieve the full network capacity and support an arbitrary mix of multi-type traffic.

For centralized wireless SDN, we develop a Mininet-based implementation of the UMW framework to evaluate its performance. In order to improve robustness in dynamic wireless environments, we modify the UMW algorithm to enable re-routing around failed links. Compared against the conventional SDN shortest path routing, our algorithm improves throughput by over 100% and significantly reduces average per-packet delay in high-throughput regime. We further present the Randomized UMW (RUMW) algorithm that performs scheduling in linear time, yet still maintains the throughput-optimality under the setting of dynamic network.

For distributed wireless SDN, our proposed Distributed Universal Max-Weight (DUMW) algorithm is throughput-optimal and non-trivially extends the UMW policy to permit distributed control and optimal inter-domain scheduling under the setting of heterogeneously delayed network state information. Furthermore, we design controller synchronization strategies that resolve the problem of multi-domain flow installation and are tailored to DUMW for maintaining throughput-optimality with negligible communication overhead. Extensive simulations validate the throughput-optimality and exhibit superior scalability of our framework.