Achieving high robustness and performance in QoS-aware route planning for IPTV networks

Quality of Service (QoS)-aware network planning in internet protocol television (IPTV) networks becomes increasingly important for network operators and Internet Service Providers (ISP) alike as different components of IPTV traffic have different and stringent QoS requirements. Proposing a routing algorithm to meet individual QoS parameters is a tedious task, since providing guarantee of meeting individual parameters in a stochastic system is impossible. Therefore, we propose optimum allocation of bandwidth where bandwidth requirements are evaluated based on accurate empirical effective bandwidth estimation for different classes of traffic by simulating a path as a First-in-First-out (FIFO) queue [2], whereby meeting the individual QoS requirements of each class of traffic. The route planning problem was formulated as a residual bandwidth optimization problem and solved using Genetic Algorithm-Variable Neighborhood Search (GA-VNS) hybrid algorithm. Although the standard evolutionary algorithms do not perform well on constrained optimization problems [4], GA-VNS was found to perform well on this particular problem [9]. This paper proposes new and novel adjustments and parameter settings to best suit the problem in terms of robustness and performance. The use of dynamically switching between two cost functions to meet the above requirements, the reduction of the difficulty of the problem in the initial generations to find a feasible solution and the use of a variation of the original VNS algorithm, besides the use of specific operators and general parameter settings are some of the recommendations of this paper. The proposed recommendations are based on extensive experiments on the performance and analysis carried out on different network topologies including Abilene topology. The proposed algorithm performed better than the recently proposed specialized constrained handling algorithms proposed in the literature, i.e. problem specific solutions are more desirable than black-box optimization for this problem. The proposed route planning mechanism was also found to produce good results even under dynamic traffic conditions.