Simulation of Failure Prone Token Rings Using Fault Injection Simulation

摘要

A simulation technique, called Fault Injection Simulation (FIS) that, amongothers, can be used for deriving waiting time distributions in failure prone token systems is presented. Token systems, and more specially token rings, are used more and more for real time applications although the influence of faults such as token losses on the system performance are fairly unknown. Although these faults are generally rare, their influence on the waiting time distribution might be important for time critical applications. For these applications there is therefore special interest in the tail in the steady state waiting time distribution, since the probability mass in the tail is a measure for how probable deadlines are missed. The basic token ring operation is modeled by a polling model. A fault occurrence and recovery model is added to this. FIS combines results of simulations of fault free systems with results of simulations of systems with moderate fault occurrence probabilities, in order to obtain results for systems with very small fault occurrence probabilities. In this way the rare event problem is circumvented. A formal framework of the technique is given and FIS is applied to an FDDI (Fiber Distributed Data Interface) token ring network as an example of its use.