Date of Graduation
12-2011
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Engineering (PhD)
Degree Level
Graduate
Department
Computer Science & Computer Engineering
Advisor/Mentor
Apon, Amy W.
Committee Member
Di, Jia
Second Committee Member
Douglas, David E.
Third Committee Member
Thompson, Craig W.
Keywords
Applied sciences; Empirical mode decomposition; Enterprise clusters; Workload characterization; Workload forecasting
Abstract
Characterization and forecasting are two important processes in capacity planning. While they are closely related, their approaches have been different. In this research, a decomposition method called Empirical Mode Decomposition (EMD) has been applied as a preprocessing tool in order to bridge the input of both characterization and forecasting processes of the job arrivals of an enterprise cluster. Based on the facts that an enterprise cluster follows a standard preset working schedule and that EMD has the capability to extract hidden patterns within a data stream, we have developed a set of procedures that can preprocess the data for characterization as well as for forecasting. This comprehensive empirical study demonstrates that the addition of the preprocessing step is an improvement over the standard approaches in both characterization and forecasting. In addition, it is also shown that EMD is better than the popular wavelet-based decomposition in term of extracting different patterns from within a data stream.
Citation
Ngo, L. B. (2011). Application of the Empirical Mode Decomposition On the Characterization and Forecasting of the Arrival Data of an Enterprise Cluster. Graduate Theses and Dissertations Retrieved from https://scholarworks.uark.edu/etd/142