An experiment using registers for fast message-based interprocess communication

Message interprocess communication has appeal for structuring operating systems but has the reputation of being slow. Register architectures have advantages for fast sequential execution but are considered to increase context-switching costs. This note describes an experiment in which registers are used for transmitting messages in the V kernel, resulting in a factor of 3 improvement in elapsed time for local message transactions and a factor of 6 reduction in kernel time for the associated kernel operations.