Design of low-cost dependable systems for distributed embedded applications.

摘要

Distributed computers systems are increasingly being embedded in complex products such as automobiles which are aimed at cost-sensitive consumer markets, and so must be both highly dependable and cheap. The overall goal of our research is to develop low-cost methods to improve the dependability of such systems.; This thesis develops software-based methods for implementing predictable and low-cost diagnosis and recovery in distributed embedded systems. We first show how to integrate time-redundant recovery policies such as task retry and primary/backup in a time triggered or statically scheduled system. Our method enables a processor to recover from transient and intermittent task failures without disrupting ongoing operations on other processors, and with moderate performance overhead, as confirmed by our experimental results. We also address distributed failure diagnosis under deadline and resource constraints and present a diagnosis scheme to identify faulty actuators in a steer-by-wire control application. Using software-implemented diagnostic tests, we formulate and solve the test-scheduling problem guaranteeing actuator diagnosis within a specified deadline while meeting the performance goals of the application. Finally, we address the design of low-cost communication networks that guarantee to meet both performance and dependability requirements of the distributed applications under consideration. We propose a design method for time-triggered communication protocols where processors use a static and global schedule to exchange messages. We also present a technique for fault-tolerant message allocation and scheduling which provides jitter-free and predictable transmission in such protocols using the minimum number of buses.