Fault-Tolerant Computing Based on Mach

Abstract

We consider the problem of providing automatic and transparent fault tolerance to arbitrary user computations based on the Mach operating system. Among the several alternatives for structuring such a system, we pursue the "task-pair backup" paradigm in detail and outline how it might be supported by Mach. Some of the new system calls and protocols that need to be incorporated into the Mach kernel and server tasks are sketched.