| Abstract: | We present algorithms for characterizing the demand behavior of applications and predicting demand by mining periodicities in historical data. Our algorithms are change-adaptive, automatically adjusting to new regularities in demand patterns while maintaining low algorithm running time. They are intended for applications in scientific computing clusters, enterprise data centers, and Grid and Utility environments that exhibit periodical behavior and may benefit significantly from automation. A case study incorporating data from an enterprise data center is used to evaluate the effectiveness of our technique.Artur Andrzejak received the PhD degree in computer science from the Swiss Federal Institute of Technology (ETH Zurich) in 2000. He is currently a researcher at Zuse-Institute Berlin, Germany. He was a postdoctoral researcher at the Hewlett-Packard Labs in Palo Alto, California, from 2001 to 2002. His research interests include systems management and modeling, and Grids.Mehmet Ceyran is working toward his Masters Degree in Computer Science at the Freie Universität Berlin, Germany. He has been employed as a student programmer at Zuse-Institute Berlin since 2003. His research interests include software engineering, systems management, and artificial intelligence. |
