Conclusion

Scalable automation of large-scale network services is a key challenge for computing systems research in the next decade. This paper deals with a software framework and policies to manage network services as utilities whose resources are automatically provisioned and sold according to demand, much as electricity is today. This can improve robustness and efficiency of large services by multiplexing shared resources rather than statically overprovisioning each service for its worst case.

The primary contribution of this work is to demonstrate the potential of model-based resource provisioning (MBRP) for resource management in a Web hosting utility. Our approach derives from established models of Web service behavior and storage service performance. We show how MBRP policies can adapt to resource availability, request traffic, and service quality targets. These policies leverage the models to predict the performance effects of candidate resource allotments, creating a basis for informed, policy-driven resource allocation. MBRP is a powerful way to deal with complex resource management challenges, but it is only as good as the models. The model-based approach is appropriate for utilities running a small set of large-scale server applications whose resource demands are a function of load and stable, observable characteristics.