An Architecture for Resource Behavior Prediction to Improve Scheduling Systems Performance on Enterprise Desktop Grids
An Enterprise Desktop Grid (EDG) is a low cost platform that scavenges idle desktop computers to run Grid applications. Since EDGs use idle computer time, it is important to estimate the expected computer availability. Based on this estimation, a scheduling system is able to select those computers with more expected availability to run applications. As a consequence, an overall performance improvement is achieved. Different techniques have been proposed to predict the computer state for an instant of time, but this information is not enough. A prediction model provides a sequence of computer states for different instants of time. The problem is how to identify computer behavior having as input this sequence of states. We identify the need of providing a architecture to model and evaluate desktop computer behavior. Thus, a scheduling system is able to compare and select resources that run applications faster. Experiments have shown that programs run up to 8 times faster when the scheduler selects a computer suggested by our proposal.
KeywordsEnterprise Desktop Grid Resource Discovery System Computer Behavior Prediction Scheduling System Classification Algorithms
Unable to display preview. Download preview PDF.
- 3.Dinda, P.: The statistical properties of host load. Sci. Program. 7, 211–229 (1999)Google Scholar
- 4.Dinda, P.: Online prediction of the running time of tasks, vol. 5, pp. 225–236. Kluwer Academic Publishers, Hingham (2002)Google Scholar
- 5.Dinda, P.: A prediction-based real-time scheduling advisor. In: Proc. International Parallel and Distributed Processing Symposium, IPDPS 2002, Abstracts and CD-ROM, pp. 10–17 (2002)Google Scholar
- 8.Kang, W., Grimshaw.: Failure prediction in computational grids. In: 40th Annual Simulation Symposium, ANSS 2007, pp. 275–282 (2007)Google Scholar
- 10.Weka Machine Learning Project. Weka, http://www.cs.waikato.ac.nz
- 11.Ramachandran, K., Lutfiyya, H., Perry, M.: Decentralized approach to resource availability prediction using group availability in a p2p desktop grid. Future Generation Computer Systems (2010)Google Scholar
- 15.Witten, I., et al.: Weka: Practical machine learning tools and techniques with java implementations (1999)Google Scholar