Violation of service availability targets in service level agreements

Targets on availability are generally included in any Service Level Agreement (SLA). When those targets are not met, the service provider has to compensate the customer. The obligation to compensate may represent a significant risk for the service provider, if the SLA is repeatedly violated. In this paper we evaluate the probability that a SLA commitment on the service availability is violated, when the service restoration time follows an exponential, Weibull, or lognormal distribution. For a two state model, where the service alternates between availability and unavailability periods, we show that such probability decreases as the variance of the restoration time grows, and that lengthening the time interval over which the service availability is evaluated reduces the risk for the service provider just if the compensation grows quite less than the length of that time interval.