GAO, LIXIN.

Since computer systems are applied to many critical areas, fault-tolerance is a necessary requirement for their operation. Many techniques for dealing with hardware faults have been developed. Fault-tolerant software has had a much slower progress. Concurrent software adds an additional dimension to the problem of fault-tolerant software. This thesis uses an intermediate structure between two major schemes, conversation and programmer transparent coordination. The scheme proposed here accelerates conversations by using a special process or superprocess, which is executed on the same system level as the run-time system, and that by having access to the history of all interprocess communications can allow a process that passes its acceptance test to proceed conditionally. If the process does not pass its acceptance test all processes recover immediately without waiting to get to their acceptance tests. This work presents a set of algorithms to implement these ideas.