Advanced stochastic host state modeling to reliable computation in global computing environment

Abstract

To design a stable global computing environment supporting reliable job execution and covering unanticipated state changes of hosts, the dynamic characteristics (i.e. volatilities) of hosts should be considered. Since a host is not dedicated to the system, voluntary hosts are free to leave and join autonomously in the middle of execution. As current systems do not relate volatility to a scheduling procedure, global computing systems suffer from performance degradation, reliability loss, job interruption, and execution time delays. For dependable computation, we propose Advanced Stochastic Host State Modeling (ASHSM), which is based on Markov model relating to execution availability quantifying duration and regularity of execution patterns of each host. Through the model, the system predicts desktop activities and allocates jobs according to the host features. Furthermore ASHSM alleviates unreliability due to unstable resource provision and job suspension during execution. © IFIP International Federation for Information Processing 2006.