Performance evaluation of abstraction models for orchestration of distributed data center networks

Cloud computing is increasingly based on geographically distributed data centers interconnected by high performance networks. Application of Software Defined Networking (SDN) is studied as an emerging solution to support dynamic network resource management for distributed data centers (DCs) jointly with extensive use of Network Function Virtualization (NFV). SDN/NFV operation takes advantage of orchestration of network control functions according to distributed DCs communication needs. Orchestration relies on a set of logical information related to the underlying infrastructure, called abstraction, which offers different levels of visibility of available resources, depending on the abstraction strategy adopted. Different abstraction strategies have been proposed in recent literature with the purpose of achieving flexible and scalable solutions for SDN/NFV operation. However, their specific application to distributed DC interconnection has not been thoroughly investigated yet. One of the critical aspects of abstraction in distributed DC operation is the amount of information provided to the orchestrator, which influences both connection request blocking probability and the rate at which orchestrator needs to update its information data base. In this paper, abstraction strategies for distributed DC networks are considered and extended with respect to previous works. Their performance are evaluated in terms of connection blocking probability and orchestrator update rate. The simulation results show that different abstraction strategies lead to different trade-offs between blocking performance and update rate. In particular, the solution based on weighted virtual links is shown to represent a flexible approach open to further possible improvement by considering different performance indicators.