Load Sharing Techniques for Server Selection in Network Virtualization

Network Virtualization enables service providers to share the same physical network infrastructure in order to create virtual networks. Service providers who own their physical infrastructure tend to maximize the revenue and resource utilization while respecting the quality of service's constraints. For this, they choose Load Sharing techniques for server selection in the context of Network Virtualization where clients send their requests and receive responses without knowing which server has responded and which path is used. Equally Distributed, Round Robin, Weighted Round Robin and Equally Distributed combined with Round Robin are proposed as Load Sharing techniques on multiple servers. In this paper, we evaluate the performance of these techniques as a function of traffic intensity in terms of latency, jitter, packet loss ratio and connection to server failure. Then, we compare these performances with those of Traffic Concentration on the closest server which is adopted by service providers who don't own their infrastructure and want to lease it. Through simulation, we notice that both Traffic Concentration and Weighted Round Robin reduce the latency when the network is slightly loaded and at medium and heavy traffics respectively. However, Equally Distributed and Round Robin provide the best performance in terms of packet loss ratio. These two simulation results are proved by an analytical model based on the M/D/1/K waiting queue. Also, the combination of Equally Distributed with Round Robin minimizes the jitter values. Finally, Equally Distributed and Round Robin decrease the probability of connection failure.