A theory for setting the "safety margin" on availability guarantees in an SLA

Abstract

A Service Level Agreement (SLA) is a contract between provider and customer that stipulates certain Quality of Service guarantees. One parameter of an SLA can be the maximum downtime guaranteed over the contract time. If the actual outage exceeds the guarantee, the customer is unhappy and the operator may bear financial penalties. It is important therefore that a network operator not only have some way to estimate or calculate the theoretical long-term availability of the services offered, but some basis for also determining a safety factor on the total outage time promised to a customer on any finite-term contract. As the period gets shorter, either nothing will go wrong and the effective availability will be perfect or if a single outage occurs, the unavailability may seem very bad. More generally, the question is how a network operator can build in theoretically justified safety factors on the availability a customer may experience over a finite-term contract. We are told that most service providers give availability guarantees based empirically on experience and historical statistics. A simple theory is thus proposed to help refine these methods for SLA policy setting. A simulation study is implemented to verify the theory within the axioms it assumes to be true.