Programming distributed fault tolerant systems

Abstract

During the last years, several toolkits for programming faulttolerant distributed systems have been proposed. Unfortunately, most of them offer a low level interface that is difficult to program and is not well integrated with a high level language. On the other hand, the Annex E of the Ada language provides a high level framework for programming distributed systems. After carefully studying it, we have concluded that nothing prevents an implementation of the Annex E that also offers support for programming fault-tolerant distributed systems. We propose to implement most of the fault tolerance mechanisms at the Partition Communication Subsystem (PCS) level, offering to the programmer the abstraction of replicated partitions, while preserving the semantics of the Annex E. Parameters related to fault tolerance can be selected at configuration time. This approach, that we have named replicAda, tries to simplify the programming of fault tolerant distributed systems by providing high level abstractions that are well integrated with the Ada language.

Multi-μ

“…ReplicAda [4] A transparent approach to fault tolerance must provide consensus in every replica's point that can lead to non-detenninistic behaviour. However, it simplifies the application programmer burden, if the fault tolerance manager can transparently implement such points.…”

Combining tasking and transaction

“…It is not surprising that there has also been various research in the faulttolerant area, such as providing recovery blocks for Ada [3], replication for Ada 83 [4], and recently even replication of Ada 95 partitions [5,6]. Support for tmnsactious in Ada 95 would be a powerful tool for a progJ'ammer of a fault-tolerant application.…”