Enabling large scale cloud services by software defined wide area network

Abstract

Interconnecting data centers (DCs) efficiently and using the fully available capacity of existing resources in Wide Area Network (WAN) seems to be one of the most challenging issues for service providers (SPs). In this master memory, we investigate a new approach to optimize traffic engineering in WAN which interconnects DCs (Inter-DC WAN) using Software Defined Networking (SDN). We propose a model to optimize bandwidth allocation to flows belonging at different Classes of Services (CoS) according to their priority and the current network state. The proposed model aims to maximize the throughput in the network and to minimize the overall energy consumption. The proposed model takes into account inter-domain communication and respects underlying technology specifications such as Multi-Protocol Label Switching (MPLS). To build our model, we consider four mathematical expressions for energy consumption of the topology nodes and links namely: the idle, the fully proportional, the agnostic and the step increasing models, and we adopt the MPLS model for Inter-DC WAN. We propose a deterministic algorithm to solve the optimization problem using Linear Programming (LP) solvers and we compare its performances with two existing models: Microsoft solutions’ SWAN which focuses on throughput maximization, and a base line model which aims to minimize energy consumption while allocating bandwidth to different flows. Experiments in the simulation environment show that the proposed solution can optimally exploit available physical capacity in the network to afford users demand in terms of bandwidth and uses the minimum energy to carry traffic. The proposed optimization model is NP-hard, so we propose a greedy heuristic to improve the runtime of the proposed solution.

Date	11 Jul 2017
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Mohamed Cheriet (Supervisor) & Kim Khoa Nguyen (Co-supervisor)