Abstract

The shared nature of the network in today’s public cloud datacenters leads to variable and unpredictable application performance. While past research has focussed on intra-tenant communication across the cloud network, we note that the emerging ecosystem of cloud applications leads to significant inter-tenant communication. This poses new challenges for cloud network sharing.

In this talk I will describe how a cloud network that carries both intra- and inter-tenant traffic should be shared. The key idea is that network allocations should be dictated by the least-paying of communication partners. This, when combined with careful VM placement, achieves the complementary goals of providing tenants with minimum bandwidth guarantees while bounding their maximum network impact. To illustrate these ideas, we have designed a prototype system called Hadrian. Results from our testbed deployment show that minimum bandwidth guarantees, apart from helping tenants achieve predictable performance, also improve overall datacenter throughput. Thus, Hadrian benefits both tenants and the cloud provider.