Impact of computational resource reservation to the communication performance in the hypercluster environment

Emerging grid computing enables researchers, scientists and engineers to build so called hypercluster, cluster of cluster, from the available resources of the grid. However, hypercluster building from the resources of the grid inherits the unpredictable resource behaviors, which result in variable performance, even using the same set of the resources but in different time. This paper proposes a design of a reservation aware operating system, RAOS, which is able to stabilize the performance of applications running on the unpredictable grid environment. We show this by stabilizing the communication cost (connection setup time, message packing time), which is one of the important parameters required to optimize in parallel applications and distributed applications, under the loaded computation node in a cluster computer.