POSER - A Process Organisation to Simplify Error Recovery

摘要

This memorandum describes a process organisation which is proposed for use in the construction of fault - tolerant, real-time computer systems. The background to this proposal is outlined in (1). The organisation is conducive to automatic checkpointing and error recovery. It is implementable on distributed computer systems, and the application programmer need not be aware whether processes to which he is communicating are in the same computer or not. The organisation comprises processes and channels which are analagous (though not identical) to those used by MASCOT (2). No shared data areas (pools in MASCOT terminology) are provided, and so communication between processes must be achieved entirely by message passing via the channels. The operation mode proposed is one of data flow at the highest level (process and channel interactions). Essentially the processes are conventional sequential programs. The channels simply move data, provide buffering, and produce the required process connectivity. This organisation is advocated as it makes the causal relations between the process and channel activations clear and thus facilitates automatic checkpointing and (backward) error recovery. The organisation may not yield very efficient implementations of certain types of application, but the efficiency has yet to be evaluated. The organisation as described here has been implemented within a multi - computer simulation in Algo168RT (3), (4) and some attempt will be made to instrument this simulation in order to try to measure efficiency. (Author)