FairShare: Dynamic Max-Min Fairness Bandwidth Allocation in Datacenters

Network is critical resource for applications, but current datacenter network is shared among untrusted tenants in a best-effort manner, which cannot guarantee fairness among tenants. Fair network sharing in datacenter is a challenging problem, because the network resource is multiplexed by many virtual machines (VMs), and every VM cannot know its explicit fair bandwidth with little knowledge of network's state. The emergence of Software Defined Network (SDN) offers us a new networking paradigm: we can maintain the network view in the controller(s), and compute fair bandwidth allocation based on the network view. In this paper, we propose an SDN based bandwidth allocation system FairShare, which efficiently provides fair bandwidth sharing. FairShare adopts Local-link Max-min Fairness algorithm (LMF) to allocate bandwidth. LMF is designed as a distributed algorithm, which is scalable and efficient for bandwidth allocation. We implement a prototype in mininet platform and a simulation system, experiment shows that FairShare can offer tenants fair bandwidths, and converge quickly to fair allocation once network turns unfair.