Detecting Violations of Service-Level Agreements in Programmable Switches

Abstract

Service level agreements (SLA) are an integral part in the delivery of services between internet service providers and their customers. They define the quality of service (QoS) that a provider promises to the customer. The implications of monitoring SLA compliance benefits both providers and customers: customers can verify whether providers uphold their QoS guarantees and providers can take corrective action if a SLA violation is detected. However, current active network measurement techniques used to monitor SLA are not accurate and are not representative of the traffic between customers and providers. In this paper, we design an algorithm and data structure to passively measure latency/RTT such that it can be applied to monitoring SLA. Our algorithm uses constant time processing per packet and our data structure uses bounded memory, making our design hardware-friendly. We evaluate the performance of our methods using trace-driven simulations on a Python prototype.