Adaptive multipath routing for large-scale layered networks

Network virtualization technologies enable multiple virtual networks to be laid over common physical networks and provide application or service-oriented networking functionalities in each virtual network. Since virtual networks share and compete for physical network resources, cooperation between virtual networks is necessary to accomplish the system-level optimization. Furthermore, system-wide adaptation is required to react to dynamically changing traffic conditions which cannot be controlled nor predicted by each virtual network. In this paper we propose an adaptive multipath routing mechanism with which globally suboptimal control can be accomplished. For this purpose, we adopt a bio-inspired algorithm, more specifically, an attractor selection model which is a nonlinear mathematical model of biological adaptation. We consider layered network architecture where a virtual network consists of virtual nodes and links which are generated by virtualizing physical domain networks and physical links. In our proposal each of physical and virtual nodes adaptively selects one of pre-established paths in accordance with dynamically changing network conditions. For cooperative routing decisions, we introduce a mechanism for routing controls operating at different virtual networks to share optimization objectives. With such loose inter-network coupling, our proposal is superior to existing control from viewpoints of adaptability and stability.