QoS-aware multipathing in datacenters using effective bandwidth estimation and SDN

Abstract

Datacenter networks are commonly structured with hierarchical topologies in which multipaths are provided to create redundant paths between end-to-end servers. Without a dedicated flow control mechanism, different sizes of traffic flows are statically allocated to links without sensing the current link utilization, which can result in transient network congestion that impacts on the latency experienced by users of the hosted applications. Existing works aim to realize load balancing by efficiently scheduling flows based on their sizes and link utilization. However, many applications' performance is bounded with specific QoS targets, so load balancing may not directly address their QoS requirements. In this paper, we present a latency-aware flow scheduling system that schedules flows based on a tenant application's QoS target and empirical estimations of the effective bandwidth required to meet these QoS targets. The approach seeks to ensure that, following a network-aware placement of VMs hosting application components, the latency associated with communications between these VMs is controlled even as the inter-VM traffic patterns change, or as the background traffic associated with other applications changes. We describe a prototype implementation of the approach that extends our MAPLE network-aware VM placement system. Our experiments in an emulated datacenter indicate that our scheduling approach leads to fewer QoS violations compared to ECMP.