We propose the new architecture to attain the following tasks: a) Optimal and deadlock-free combination of global load balancing with dynamic task scheduling on a multiprocessor machine, yeilding busyness close to one for arbitrary algorithm; b) Optimal and deadlock-free fault tolerant execution with the minimum number of messages and no postponements of the underlying computations; and c) Assurance of responsiveness and timeliness by integration of our optimal task management with our optimal fault-tolerant computing in such a way, that they support each another.
Keywords: