The design of a fault management framework for cloud

Abstract

High performance computing systems can have high failure rates as they feature a large number of servers and components with intensive workload. The availability of the system can be easily compromised if the failure of these subsystems is not handled correctly. This research proposes a framework of proactive fault tolerance for enterprise cloud computing systems. The main idea is to create an effective prediction model focusing on hardware failure. The proposed framework features two major components: monitoring and availability analysis. For each machine, the availability analysis module tracks historical states, and predicts the machine future state. Depending on the predicted state, the resource manager decides whether the machine requires task migration to prevent possible losses. By using task migration, the framework eliminates the cost of job replication and back up. The framework also includes the adequacy checking function into availability analysis in order to periodically evaluate and adjust the prediction model. The framework can thus be adopted by heterogeneous datacenters. The energy efficiency can be improved as the impact of the failure to the datacenters reduces.