Enhancing the performance and dependability of real-time systems

The imprecise-computation technique was proposed as a way to handle transient overloads and enhance the dependability of real-time systems. In a system based on this technique, each time-critical task produces an approximate but usable result whenever a failure or overload prevents the system from producing a precise result. This approach makes meeting deadlines easier, increases the availability of data and services, reduces the need for error-recovery operations, and minimizes the costs in replication. We provide an overview of ways to implement and schedule imprecise computations. We then describe the Imprecise Computation Server (ICS) and how it is used to ensure real-time performance of time-critical applications. >