Optimal traffic routing in the network virtualization context

Abstract

Network virtualization enables service providers to instantiate virtual networks on a common physical infrastructure. Virtual networks are then allocated to clients in order to deploy or test their services. Thus, clients do not know how their traffic is routed, which server is responding, and which path is crossed. Service providers who own the physical infrastructure aim to share the load among multiple servers for traffic routing. However, those who do not own the infrastructure and want to lease it essay to concentrate all the traffic on one server and exactly the closest server. In this paper, we model the problems of load sharing on multiple servers and traffic concentration on one server as two mathematical programs. The objective functions correspond to service providers' aims under a delay constraint based on the M/D/1/N queue's mean sojourn time's formula. Due to its hardness, we provide an approximation of the delay constraint at low and medium traffic intensities. The approximation is analytically proved by the Taylor series development at a chosen point. Finally, through simulation and analytical models, we compare the performance of programs' optimal solutions. We notice that load sharing decreases the packet loss and jitter values. However, traffic concentration reduces the latency.

Multiobjective Path Problems and Algorithms in Telecommunication Network Design—Overview and Trends

“…Routing in these networks is controlled by the service providers that manage the transport infrastructure. A routing-related problem is the need to guarantee traffic load sharing over multiple paths (thus leading to multipath routing) among different servers (from the perspective of the service provider) or to concentrate the traffic on one particular server (from the perspective of the clients with virtual networks) [207].…”