Design of Fault Tolerant Software

In this paper we deal with structured software fault-tolerance. Structured software fault tolerance are those techniques where redundancy (both for detection and correction) is applied to the individual blocks of software with the goal of masking or reveal errors internal to the block. Each technique has its own way of structuring the interactions among redundant parts and of managing the complexity added. We discuss some of the open problems of software fault tolerance structures and the issues related to effectiveness. In particular we address generality and flexibility, which can be improved at the price of adding complexity to the design, and discuss the need of proper trade-offs between generality and flexibility on one hand and complexity on the other. The SCOP (Self-Configuring Optimal Programming) scheme [7, 24] constitutes an interesting example of such a trade-off. SCOP is a fault tolerant system structuring method, originally intended to improve run time efficiency, that is also very flexible and general. Together with the mechanisms used by SCOP for obtaining flexibility and generality, some recent research aiming at reducing the complexity is finally described