Online Scheduling for Cloud Computing and Different Service Levels

In this paper, we address scheduling problems for infrastructure as a service (IaaS). In a typical IaaS scenario, an infrastructure provider offers his resources on demand and with different service levels to his customers. These service levels are mainly distinguished by the amount of computing power a customer is guaranteed to receive within a time frame. In our a model, each service level is described by a slack factor and a price for a processing time unit. If the provider accepts a job it is guaranteed to complete by its deadline, that is its submission time plus its processing time times the slack factor of assigned service level. After a job has been submitted, the provider must decide immediately and irrevocably whether he accepts or rejects the job. We suggest various algorithms and use competitive analysis to discuss different scenarios for this model. These scenarios combine fixed services levels with the single machine model or the parallel identical machines model. Particularly, we demonstrate the benefit of parallelism by showing that we can achieve better competitive factor in a parallel machine scenario than in the corresponding single machine scenario.