Decomposition of object-oriented programs for fault tolerant computing in distributed environment

This paper describes a program decomposition scheme that decomposes a C++ program into a set of concurrent tasks to support fault tolerant computing in distributed environments. This scheme consists of a C++ analyzer for program decomposition and a set of mechanisms to perform run-time object backup and error recovery. The analyzer uses a weighted-object approach to decompose an object into a set of concurrent tasks or combines a set of objects into a single task. The analyzer guarantees that the resulting decomposition will not violate system resource constraints. A reliable message passing protocol is also generated by the program analyzer for remote object references. Run time supports for fault tolerant computing based on the decomposed objects are discussed, which consist of an interface, a reliable message passing mechanism and a recovery mechanism. The message passing mechanism updates backup storage whenever a message is passed to avoid Domino effect.<
>