Flexible schemes for application-level fault tolerance

This report considers application-level fault tolerance, that is, fault tolerance policies explicitly programmed in the upper layers of a (hardware plus software) computing system. Design of fault tolerance at this level is normally necessary. but difficult and error-prone due to its mostly ad-hoc nature. There is a need for discipline to control the complexity added to any design by the provision of fault tolerance, and to allow the automation of some part of the task of redundant design. At the same rime, the structuring schemes proposed in the literature appear too restrictive for the building of large, heterogeneous applications. Ways to overcome these limitations while preserving a structuring discipline are discussed first with respect to the detection and signalling of errors, and then, to a wider extent, to recovery.