Implications of Recovery Schemes for Virtualization Platform

Abstract

In this paper, we propose a general simulation platform integrated with a general failure model framework. Based on our simulation platform, we investigate the implications of system failures on virtualization platform. Meanwhile, we investigate the capability of virtualization platform's recovery mechanism, both reactive and proactive ones. In fact, we find that proactive recovery, in terms of both performance and resource debts, outperforms reactive ones. It has the potential to extract the most of performance gap between system with failures and one without. Furthermore, as proactive recovery depends on the results of system status predictor, we investigate the impacts of predictor's attributes and find that predictor's 'False Negative' attribute is most important for exploiting the proactive recovery mechanism's capability.