On the feasibility of automation for bandwidth allocation problems in data centers

Abstract

Mapping virtual networks to physical networks under bandwidth constraints is a key computational problem for the management of data centers. Recently proposed heuristic strategies for this problem work efficiently, but are not guaranteed to always find an allocation even when one exists. Given that the bandwidth allocation problem is NP-complete, and the state-of-the-art SAT solvers have recently been successfully applied to NP-hard problems in planning and formal verification, the goal of this paper is to study whether these SAT solvers can be used to solve the bandwidth allocation problem exactly with acceptable overhead. We investigate alternative ways of encoding the allocation problem, and develop techniques for abstraction and refinement of network graphs for scalability. We report experimental comparisons of the proposed encodings with the existing heuristics for typical data-center topologies.