SLA learning from past failures, a Multi-Armed Bandit approach

A Service Level Agreement (SLA) is a contract between a customer and a Network Service Provider (NSP) defining services to one or more destinations at a given Quality of Service (QoS). Once committed, the SLA can be violated without the customer being able to predict that. We focus on offer selection mechanisms according to QoS and taking into account past SLAs´ violations. We propose an algorithm, using a minimizing-regret technique, which provides an estimation of the reliability of given NSPs to the customer. Our approach only requires an end-to-end monitoring tool for used paths and depends on the customer´s history.