Network virtualization on the wireless edge

Abstract

This thesis presents a comprehensive investigation of wireless network virtualization, a technique for creating multiple independent software-definable networks on a single set of commercial hardware resources. Network virtualization has previously been applied to wired networking scenarios, but the general problems of wireless virtualization represents an important open problem that we address in this work. In particular, we identify key technical challenges, system concepts and architectures, as well as specific protocols and algorithms for implementing wireless network virtualization. In summary, this thesis will provide results for following aspects of wireless network virtualization: (1) Basic mechanisms for link (spectrum) sharing and their isolation performance with virtual WLAN networks, (2) Virtualization mechanisms and traffic isolation algorithms for virtualized WiFi networks, (3) Virtualization of cellular basestations including experimental evaluation for a prototype 4G/WiMAX network, and finally, (4) analytical evaluation of virtualization algorithms for more general multi-hop wireless topologies. The first part of the thesis presents an exploratory discussion on the co-existence of multiple 802.11 based virtual networks. A comparison is presented for understanding the tradeoffs between sharing the radio through spatial and temporal separation on the ORBIT wireless testbed. Experimental evaluations reveal that while virtual networks sharing channel resources by space separation achieve better efficiency than those relying on time, the isolation between experiments in both cases is comparable. Supporting virtualized WiFi access point based networks allows for a convenient sharing of a physical access point across multiple ISPs or network operators. The second part of the thesis discusses our SplitAP architecture, which builds on the virtual access point (VAP) mechanism by extending it to support fair-sharing of airtime across multiple wireless networks. This is done by implementing a dynamically controlled isolation framework across competing slices. The framework also allows the user to deploy custom algorithms for enforcing uplink airtime fairness across client groups within the SplitAP framework. The thesis shows up to 40% improvement in isolation measured through a modified Jain fairness index with LPFC and LPFC+, two sample algorithms implemented on the framework. The third part of the thesis addresses the challenge of virtualization of resources in a cellular basestation (BTS) while allowing operators to use distinct flow types, quota allocations, slice schedulers, and network layer protocols. The proposed virtual basestation architecture is based on an external substrate which uses a layer-2 switched datapath, and an arbitrated control path to the WiMAX base station. The virtual network traffic shaping (VNTS) slice isolation mechanism allows the virtual basestation users to obtain at least an allocated percentage of the BTS resources in the presence of saturation and link degradation, helping make the performance repeatable. Performance measures such as fairness index and coupling coefficient are defined and evaluated experimentally, showing significant improvements with preliminary indoor mobility experiments. Outdoor vehicular measurements show similar improvements in the fairness index and coupling coefficient, demonstrating the feasibility of the proposed VNTS algorithms. Finally, a theoretical formulation describes how a mapping mechanism can be used for provisioning and allocating resources on wireless networks that are supported by wireless virtualization schemes such as the virtual basestation and the SplitAP framework. Results show that the wireless mapping problem can be reduced to solving a combinatorial optimization problem at nodes selected greedily based on their capabilities to generate revenue. Detailed simulations are discussed for highlighting the performance of the proposed greedy static allocation (GSA) and greedy dynamic re-allocation (GDR) algorithms.