The authors demonstrate the need to address fault latency in highly reliable real-time control computer systems. It is noted that the effectiveness of all known recovery mechanisms is greatly reduced in the presence of multiple latent faults. The presence of multiple latent faults increases the possibility of multiple errors, which could result in coverage failure. The authors present experimental evidence indicating that the duration of fault latency is dependent on workload. A synthetic work generator is used to vary the workload, and a hardware fault injector is applied to inject transient faults of varying durations. This method makes it possible to derive the distribution of fault latency duration. Experimental results obtained from the fault-tolerant multiprocessor at the NASA Airlab are presented and discussed.>

It is shown how to determine closed-form expressions for task scheduling delay and active task time distributions for any real-time system application, given a scheduling policy and task execution time distributions. The active task time denotes the total time a task is executing or waiting to be executed, including scheduling delays and resource contention delays. The distributions are used to determine the probability of dynamic failure and processor utilization, where the probability of dynamic failure is the probability that any task will not complete before its deadline. The opposing effects of decreasing the probability of dynamic failure and increasing utilization are also addressed. The analysis first addresses workloads where all tasks are periodic, i.e., they are repetitively triggered at constant frequencies. It is then extended to include the arrival of asynchronously triggered tasks. The effects of asynchronous tasks on the probability of dynamic failure and utilization are addressed.>

A closed queueing network model is constructed to address workload effects on computer performance for a highly reliable unibus multiprocessor used in real-time control. The model consists of multiserver nodes and a nonpreemptive priority queue. Use of this model requires partitioning the workload into task classes. The time-average steady-state solution of the queuing model directly produces useful results that are necessary in performance evaluation. The model was experimentally justified with the fault-tolerant multiprocessor located at the NASA AIRLAB. Extensive experiments were performed using a synthetic workload generator to measure directly performance parameters, such as processor idle time, system bus contention and task processing times. These measurements determine values for parameters in the queueing model. Experimental and analytic results are compared, demonstrating the validity of the model.>