Achieving bandwidth guarantees in multi-tenant cloud networks using a dual-hose model

Abstract

In public cloud networks, applications of different tenants compete for the shared network bandwidth and thus might suffer from unpredictable performance. It is desirable for cloud providers to offer tenants with bandwidth guarantees. However, it is challenging to precisely abstract tenant bandwidth requirements for their intra- and inter-tenant communications and to achieve work conservation simultaneously. In this paper, we first propose a dual-hose model, a novel tenant requirement abstraction that decouples bandwidth guarantees for a tenant's inter-tenant communications from those for its intra-tenant communications. We then develop a new VM placement algorithm to optimize operational goals of cloud providers, while providing tenants with minimum bandwidth guarantees captured by the dual-hose model. Finally, we design a dynamic bandwidth allocation strategy to achieve work conservation. Through extensive simulation results, we show that our solution provides bandwidth guarantees for tenant requests while improving the overall request throughput by 5.3%.