The imprecise computation technique has been proposed as a way to hand
le transient overload and to enhance fault tolerance of real-time syst
ems. In a system based on this technique, each time-critical task is d
esigned in such a way that it can produce a usable, approximate result
in time whenever a failure or overload prevents it from producing the
desired, precise result. This paper describes ways to implement impre
cise computations, models to characterize them and algorithms for sche
duling them. An imprecise mechanism for the generation and use of appr
oximate results can be integrated in a natural way with a traditional
fault-tolerance mechanism. An architectural framework for this integra
tion is described.