A control-theoretic approach to automated local policy enforcement in computational grids

A computational grid is composed of resources owned and controlled by a number of geographically distributed organizations. Each organization demands site autonomy — each organization must have complete, direct control over their resources, e.g., regulating how much resource can be used, by whom, and at what time. One of the reasons for site autonomy is to ensure that local computation jobs within the organization can obtain the resources in a timely fashion. Typically, policy enforcement lies on the shoulders of the system administrators in each organization. We see there is a great need for developing a new automated method to facilitate this process. In recent years, there has been a great deal of research on grid economies to provide accounting management and economic incentives for allocating and sharing grid resources. In this paper, we describe a control-theoretic approach for automated enforcement of local policy built upon the notion of computation grid economy. A closed-loop system is constructed by adding a controller that manipulates the resource price using a proportional, integral control law. Our simulation results show that the controller helps the system to quickly achieve the targeted utilization and adapt to changing conditions.