A Modified Adaptive Space-Sharing Policy for Non-dedicated Heterogeneous Cluster Systems

Adaptive space-sharing algorithms are commonly used to realize the fullest strengths of dedicated homogeneous cluster computing systems since they produce better performance than fixed and variable-space sharing policies. However cluster of computers tend to be heterogeneous over the time because of both nature of incremental growth and diversity of technological change. Community-owned clusters are not only heterogeneous but are also non-dedicated i.e. computational resources are shared between local and parallel workload. The existing adaptive policies for dedicated cluster systems are not suitable for such conditions. This paper modifies an existing adaptive policy for dedicated heterogeneous systems so that it can be adapted to work in non-dedicated heterogeneous systems. Evaluation results show that the proposed algorithm provide substantial improvement over existing similar algorithms at moderate to high system utilizations.