Reliable Management of Virtualized Resources Using Fault Trees

The new trends in distributed computing has changed the way we do computing when talking about cloud infrastructures or high-performance computing. Resource virtualization technologies enabled elasticity of resource provisioning and management through easy replication of virtual nodes or virtual machine migration. In order to provide high availability and reliability in such distributed environments where resources are managed and served in form of virtual machines, specific load balancing and fault strategies are needed. Based on fault tree analysis concepts, we propose a distributed and autonomous approach to manage faults using fault agents able to asses and predict for each virtualized node, its state of fault or future fault. Accordingly, each node can take a decision about accepting future jobs, delegate jobs to own replicated instances or start a live migration process as a second strategy for assuring availability and continuity of the service.