Fail-Safe PVM: A Portable Package for Distributed Programming with TransparentRecovery

摘要

Through fail-safe PVM we seek to explore fault tolerance for distributedcomputation from a practical perspective. The fail-safety features are considered to be add-ons to an existing model for distributed computing, VIZ PVM in the prototype. We desire the following from the add-ons, with respect to the base model: (1) Application Independence. Fail-safety is provided for arbitrary model programs; (2) Application transparency. Failures are invisible to applications that do not bypass the model; (3) Compatibility. The interfaces presented by the model and the modified model are compatible; (4) External to operating system. The implementation requires only the standard OS interface. This makes it portable, which in turn facilitates heterogeneity and the constitution of systems containing many machines from different administrative domains; (5) Minimal overhead. Minimize overhead during regular execution; and (6) Practicality. The whole system is considered. In particular, the existence of distributed stable storage is not assumed.