System mechanisms for distributed object-based fault-tolerant computing

A scheme for comprehensive fault tolerance must address both consistency and forward-progress requirements of applications. We propose one such novel scheme for distributed applications based on the shared memory paradigm and structured using objects. Since no fixed set of existing transaction models meets the requirements of all applications, our scheme provides system mechanisms that can be used, directly or, to implement generalized and replicated transaction models that correspond to policies for fault tolerance. We demonstrate how our mechanisms can be used by implementing the nested transaction model. We have implemented the mechanisms in the Clouds distributed operating system and present a performance study of the system primitives