In simple words, cloud customers have different demands, cloud infrastructures have HW/SW heterogeneity, and failure is inevitable. So why are the providers provisioning fixed downtimes guarantees for everyone?
This paper introduces a setting where clients can ask their desired availability and get charged correspondingly. We use game theory to build an incentive compatible system where everyone wins. We also implement a prototype of this system in OpenStack and show that dealing with flexible availability is not as hard as it sounds. Applying this system, without any hardware change, can decrease the provider cost by 10%, increase its profit margin by 20%, and more importantly improve the overall user satisfaction.