Run-time models for dynamic resource management in distributed systems

The success of Web related technologies is leading to the implementation of large networked systems that support heterogeneous, possibly critical, services. Dynamic capacity management of these services would increase efficiency and business productivity, but similar processes are difficult to be implemented in contexts characterized by extreme variability of workloads and resource states to the extent that it is difficult to identify stationary periods. We propose an innovative multi-phase methodology that is based on stochastic representations of the resource measures and on accurate models for positioning the present and future state of the system resources with respect to their capacities. These models represent the basis of several runtime decision algorithms that are oriented to load balancing, access control, anomaly detection, energy saving. This research area is characterized by various open issues that will be outlined at the end of the talk.