Robust Resource Allocation Techniques on Homogeneous Distributed System

Abstract

Distributed computing systems utilize various resources with different capabilities to satisfy the requirements of diverse task mixtures and to maximize the system performance. Such systems often operate in an environment where certain desired performance features degrade due to unpredictable circumstances, such as higher than expected work load or inaccuracies in the estimation of task and system parameters. Thus, when resources are allocated to tasks it is desirable to do this in a way that makes the system performance on these tasks robust against unpredictable changes. The system is considered robust if the actual makespan under the perturbed conditions does not exceed the required time constraint. The goal is to maximize the collective allowable error in execution time estimation for the tasks that can occur without the makespan exceeding the constraint.