An efficient and lightweight method for Service Level Agreement assessment

Traditional approaches to on-line end-to-end Service Level Agreement (SLA) assessment have focused on the estimation of network QoS parameters. These approaches, however, face a trade-off between accuracy and the amount of resources needed to achieve such accuracy. This paper offers an alternative approach, where instead of estimating QoS parameters, we propose an effective and lightweight solution for directly detecting SLA violations. Our solution monitors the Inter-Packet Arrival Time (IPAT) at an end-point, wherein current IPAT distributions are periodically compared with a set of reference IPAT distributions as the main basis for detecting SLA violations. A mapping of the IPAT distribution with the current network conditions is derived, and a training algorithm that dynamically acquires the set of reference IPAT distributions is designed. For the comparison of the IPAT distributions, we propose a variant of the Hausdorff Distance algorithm. Our variant provides a better accuracy than the traditional Hausdorff Distance, while presenting linear complexity. Our proposal is validated in a real testbed, by comparing the SLA violations detected and the resources required in terms of bandwidth, with other existing alternatives as well as with perfect knowledge of current network QoS status.