Routing algorithm for multiple unsplittable flows between two cloud sites with QoS guarantees

Abstract

Large scale cloud applications may require users to get multiple resources (e.g., VMs, storage) from different sites and simultaneously connect each pair of resources by a path that can satisfy certain Quality-of-Service (QoS) requirements. Finding such paths with the bandwidth constraint is known as the unsplittiable flow problem, which is shown to be NP-hard. Accordingly, various approximations as well as heuristic algorithms are proposed to maximize the amount of routed flows. However, the existing solutions suffer from either low performance in maximizing the amount of routed flows or excessive computation time. In this paper, we propose a new efficient heuristic for a special case, where multiple flows are sent between two cloud sites. Our heuristic algorithm first determines level cuts, set of links that are at same distance from source to destination, whose removal disconnects the source and destination. It then uses randomized best fit algorithm for assigning the flows to the links in the possible bottleneck cuts, which are the level cuts whose total available bandwidth is less then (1+γ) times the total requested bandwidth. It finally tries to forward flows between the level cuts using bandwidth constrained shortest path algorithm per flow. We demonstrate the efficiency of our heuristic using simulation.