Kernel services approach to fault-masking in real-time applications

The N-Modular Redundancy (NMR) architecture has shown to perform fault-masking with acceptable cost and performance. The restriction to the majority voting imposed by many of its implementation models often discourages its use in real-time applications with sensors' data. Its implementation in real applications also exhibits time overhead, and synchronization problems beside the increase in hardware cost it involves. This paper shows a way to cope with these problems by using a distributed run-time kernel in which each service is offered in two versions, a non-NMR version and an NMR one. This offers the user a transparent way of developing its real-time fault-masking applications.