Replicated processing with voting provides a powerful means of constructing highly reliable computing systems. We will consider a functionally distributed computing system intended for real-time applications, where each functional module — a node — has been configured in an NMR (N-modular redundant) fashion. Such a system receives processing requests from ‘actuators’ (the entities that demand services) that require distributed processing at various nodes. The paper will discuss various approaches to scheduling computations to ensure that each processor of an NMR node processes input messages in an identical order. The concept of exception handling for voters will be developed to detect failures in the system.