On revenue driven server management incloud

As failures are becoming frequent due to the increasing scale of data centers, Service Level Agreement (SLA) violation often occurs at a cloud provider, thereby affecting the normal operation of job requests and incurring high penalty cost. To this end, we examine the problem of managing a server farm in a way that reduces the penalty caused by server failures according to an Infrastructure-as-a-Service model. We incorporate the malfunction and recovery states into the server management process, and improve the cost efficiency of server management by leveraging the failure predictors. We also design a utility model describing the expected net revenue obtained from providing service. The basic idea is that, a job could be rejected or migrate to another server if a negative utility is anticipated. The formal and experimental analysis manifests our expected net revenue improvement.