Novel Cluster Performance Matrix: A Decision Aid for Evaluating Computing Hardware
Sarmad Alshawi, Zahir Irani, Ayad Jassim · Journal of Intelligent Systems · 2000
Many organizations are increasing their expenditure on InformationTechnology (IT) to obtain or even to sustain a competitive advantage in their respective marketplaces.Nevertheless, managers are often left with the quandary of how to evaluate investments in IT hardware/software.Reasons for this difficulty have been suggested by the normative literature as centering on the socio-technical dimensions associated with IT adoption.The inability of managers to select the appropriate computing hardware are considered attributable to a lack of appropriate decision aids that might act as a framework culminating in knowledge and understanding about the true performance of the IT.In developing a broader understanding of the need for a decision aid to support the evaluation of computing hardware clusters, the authors of this paper introduce a novel model for presenting performance data, thereby supporting IT specialists in the evaluation of computer clusters.The need for such a novel decision aid is justified as computing clusters can vary significantly in terms of their hardware and software configurations.In addition, there may be several combinations in which a parallel job can allocate its threads among cluster nodes and the number of nodes used.Such diversity makes the process of comparing cluster performance a very complex procedure.The model presented in this paper is based upon the notion of a 'Cluster Performance Matrix' and its derived metrics.A numerical example, based on a computational fluid dynamics application, will be used to 539 Vol. 10.No. 5-6.2000 Novel Cluster Performance Matrix demonstrate the applicability of the underlying model when using two different hardware clusters.