A Service-oriented Measurement Infrastructure for Cloud Computing Environments

Abstract

Abstract-With an ever-increasing number of services being moved to cloud computing platforms, it is essential for data centre operators to quickly identify any problems in their network in order to ensure optimal performance and provide guidance for when actions such as migration of computing instances are required. This paper introduces a distributed measurement infrastructure using low-cost in-line link measurements to reveal the true traffic-perceived service quality in short timescales. In-line measurement is highly relevant for data centres as it allows an accurate measure of important network services without the problem of injecting artificial traffic into a network with high contention for resources. The infrastructure presented allows real-time monitoring and visualisation of the state of the network, which can be used to inform virtual machine migration/instantiation events. We show that the in-line measurements used by the infrastructure have negligible impact on network traffic and the hosts conducting the measurements. The low overhead of in-line measurements potentially enables the infrastructure to function as an always-on measurement and management platform.