Resource Access Pattern Mining for Dynamic Energy Management

Energy management mechanisms in resource constrained environments such as mobile systems are typically based on the current status of various system parameters such as utilization and quality of service requirements. Any knowledge gained about future system resource usage and other significant dynamics in the system can further lead to increased optimizations in such mechanisms. Future knowledge about system resource utilization and dynamics can be derived from two separate sources: a) the information maintained in the system about its current utilization and resource accesses and, b) the application specific behavioral patterns of currently active applications. This work targets at employing known efficient data mining techniques to exploit the latter class of information in achieving a suitable application behavior model that can provide accurate predictions of future application specific resource accesses and usage. Such formulated application behavioral patterns containing information about their resource usage, termed Resource Access Patterns, are then employed in a novel dynamic energy management scheme for a mobile system network resource. This work also shows that a data mining approach is beneficial in energy management decisions as it can capture all the underlying intricate interactions and dependencies that exist among the active applications and between the resources available in the system. The key components of this work can thus be summarized as: a) employing a data mining approach to characterize application behavior and, b) designing a novel energy management policy for a network interface device based on information about its future usage derived from the formulated application behavioral model.