Skip to main content

Application of Support Vector Machine approach to Chinook failure prediction


EMSL Project ID
33195

Abstract

Mitigating the impact of computer failure is possible if accurate failure predictions are provided. This research will provide a proof of concept implementation of a new spectrum-kernel Support Vector Machine (SVM) approach to predict failure events based on system log files and other data. Experimental results using actual system log files from the MPP2 Linux-based compute cluster indicate the proposed SVM approach can predict hard disk failure with an accuracy of 76% as much as 48 hours in advance. We propose to apply similar analysis to Chinook, developing new techniques and studying the effects of a larger-scale system. The payoff for this research will be increased reliability of HPC systems and less expense from time wasted due to component failures.

Project Details

Start Date
2009-01-26
End Date
2012-01-29
Status
Closed

Team

Principal Investigator

David Cowley
Institution
Environmental Molecular Sciences Laboratory

Related Publications

Fulp EW, GA Fink, and JN Haack. 2008. "Predicting Computer System Failures Using Support Vector Machines." In First USENIX Workshop on the Analysis of System Logs (WASL '08). USENIX, Berkeley, CA.