Data Plane Traffic Engineering in Multimedia Communications using Software-Defined Networking

The quality of multimedia communications heavily relies on the condition of the end-to-end network. Sub-optimal resource allocation may deplete certain links sooner, making the network less resilient to link failures, traffic fluctuation or random traffic spikes. Regulating traffic can improve the media quality, but is not an easy operation to do in legacy IP networks. The maturing Software Defined Networking (SDN) technology provides the support for an unprecedented centralized solution in this regard: making it possible to deploy the optimal embedding of data flows via the central SDN controller. Thus the biggest challenge is to calculate such an optimal embedding for the data flows. This thesis presents our design of traffic engineering in multimedia communications which addresses to the resource allocation concerns in the data plane. Given the physical network, our objective is to allocate the network resource and embed the multimedia data flows in the way such that the total throughput is maximized. To achieve this, we introduce a Centralized Data Flow Routing Module into the control plane of the system. The Module employs a Two-Phase flow embedding procedure: Phase \RNum{1} is a global planning subroutine, periodically invoked. Given the traffic input of an extended time period, it produces and installs the optimal end-to-end tunnels between each pair of communication endpoints. Phase \RNum{2} is a Traffic Engineering (TE) subroutine, always active. It receives the individual data flow requests, and selects one of the pre-installed tunnels to embed the flow. For Phase \RNum{2}, first we propose a greedy-based TE algorithm. We investigate its performance and compare it with the performance if a Dijkstra-routing algorithm is employed. We show that the proposed heuristics outperform the Dijkstra's in three different metrics: session accept rate, maximum/average physical link utilization, and network criticality. In a second attempt, we present an incremental TE algorithm for Phase \RNum{2}. We numerically evaluate it and compare it with (\RNum{1}) Dijkstra algorithm and (\RNum{2}) the greedy-based TE heuristics. We show that the incremental TE excels in all three metrics: session accept rate, throughput, and link utilizations, under a wide spectrum of various network conditions and traffic conditions.Ph.D