Abstract: Prediction and Adaptation in Active Harmony

Active Harmony is a new infrastructure for managing resources in large, dynamic environments. One of Harmony's major strengths is its ability to gather and exploit many types of application-specific informa-tion. This paper concentrates on two. First, the LBF metric predicts the performance impact of moving procedures and processors. Second, active correlation-tracking is used to gather data sharing information and drive thread migration decision heuristics.

However, Harmony's strength is not in the individual mechanisms and metrics used, but in the interfaces that allow global policy decisions and many types of applica-tion-specific information to be tied together. Harmony is a work in progress, and we are continuing to evaluate new sources of information and new mechanisms for possible inclusion in the system.