Importance Sampling to Evaluate Real-time System Reliability: A Case Study

Real-time distributed computers are often used in life- critical applications. However, the complexity of such systems calls for extensive simulation studies to validate their performance and reliability before a design can be accepted and a prototype constructed. A simulator testbed has been built to model a variety of such systems quickly from a few basic building blocks. Life-critical applications require reliability levels so high that brute-force simula tion to validate these levels would take weeks of computer time. In this paper, we present studies we have conducted into the use of importance sampling in simulating real- time systems. This paper presents a interesting case- study of the use of importance sampling in an increas ingly important branch of computer engineering. Impor tance sampling may not work for all cases and over all parameter ranges. In this paper, we are interested in finding out whether (and how well) this scheme works for the case of distributed real-time systems and also the range of failure bias values for which it works well. Specifically, we look at the implementation of two heuris tics called 'forcing' and failure biasing' in the testbed. This was validated by comparing the reliability estimates with that of normal (very long) simulation. The effect of the failure bias on the dynamics of the scheme is also investigated to provide readers with some guidance on choosing appropriate bias values.