Approaches to cloud computing fault tolerance

Fault tolerance in cloud computing platforms and applications is a crucial issue. This issue is especially difficult since cloud computing relies by nature on a complex splitting into many layers. From a technical point of view, most cloud computing platforms exploit virtualization, which implies that they are split into 3 layers: hosts, virtual machines and applications. From an administration point of view, they are split into 2 layers: the cloud provider who manages the hosting center and the customer who manages his application in the cloud. This paper analyses the implementation of fault tolerance in such a complex cloud computing environment with a focus on autonomic repair. We show that in most of current approaches, fault tolerance is exclusively handled by the provider or the customer which leads to partial or inefficient solutions, while collaborative solutions are much promising. We illustrate this discussion with experiments where exclusive and collaborative fault tolerance solutions are implemented in an autonomic cloud infrastructure that we prototyped.