Three-layer control policy for grid resource management

The paper proposes a three-layer control policy for grid resource management, which applies a bottom-up approach to dynamic resource management. In order to hierarchically allocate resources in the system to maximize system utility, different controllers are deployed at three levels: local controller, group controller, and global controller. Global control considers all applications and coordinates three layers of grid architecture in response to large system changes at coarse time granularity. While local control adapts a single application to small changes at fine granularity. Global control exploits the interlayer coupling of fabric layer, collective layer and application layer to perform less frequent and effective control actions for a system-wide optimization. The interaction between grid layers is controlled through the use of the pricing variable, which coordinates the user demand at the application layer and supply of resources at the fabric layer. The paper proposes pricing-based three-layer control algorithms. This paper demonstrates the benefits of the control algorithm through simulations.