MANAGING ON-DEMAND COMPUTING SERVICES WITH HETEROGENEOUS CUSTOMERS

Cloud computing allows the practice of allocating shared computational resources to a diverse set of customers in an on demand fashion. The heterogeneity of customers' requirements can be characterized by the frequency of demand, the value associated with the service and delay sensitivity. We consider the following two classes of customers: frequent users (FC) and urgent users (UR). The FC agreement is a proxy for the current business model used by typical providers; all customers are guaranteed the service for a flat, usage-based fee. Any delays are subject to penalties in FC. In the UR, service is guaranteed to be completed without any delay, but the service provider can accept/reject UR job requests with no penalties. We focus on the admissions control decision for the UR customers. Using computational experiments and simulation, we test the effectiveness of heuristic admissions control policies.