Joint load balancing and energy saving algorithm for virtual network embedding in infrastructure providers

Abstract

Network virtualization is key to cloud services, in that it enables multiple users to share a physical infrastructure through abstraction. We propose an online virtual network (VN) embedding scheme which jointly considers load balancing and energy saving so as to maximize the profit of Infrastructure Providers (InPs). For load balancing, we propose to minimize a convex objective which penalizes mapping of VNs to overloaded resources. For energy saving, we consider two popular energy models: speed scaling and power-down. In the speed scaling model, energy consumption is modeled as a convex function of the load imposed on resources. We observe that both load-balancing and energy-saving objectives superadditively penalize high utilization/congestion at resources, and that such synergistic nature of the objectives leads to efficient joint optimization. In the power-down model, a fixed cost exists for keeping a node powered on, which is characterized by a nonconvex energy curve. In this case, we propose an iterative algorithm which explores the trade-offs between load balancing versus cost reduction from power-down of idle servers, in a controlled way. Our algorithm performs a sequential node and link mapping; in particular, for link mapping, we adopt randomized rounding with path stripping in order to obtain a constant factor approximation to the minimum penalty for link utilization. Numerical experiments show the efficacy of our algorithm in servicing VN requests of various topologies and resource requirements.