Virtual Duplex Systems Provide the Cost Benefit of Requiring Only One Rather than Two Processors . to Lighten the Single Processor ’ S Burden and Meet Real - Time Requirements

Embedded systems play an important role in our daily lives: We interact, sometimes unknowingly, with embedded systems in applications ranging from cars, trains, and aircraft to washing machines and other household appliances. In an increasing number of applications, the system’s ability to detect or tolerate transient and permanent faults is important. To achieve this goal, a system requires redundancy, or multiple resources. In the past, designers have used some form of duplex system, or structural redundancy, for these applications. However, the cost associated with duplicated hardware resources is a major drawback of duplex systems. For embedded systems, which are usually deployed in very high volumes in highly competitive products such as household appliances, there is a strong motivation to lower costs. In this article, we present a new approach, which avoids the high costs associated with traditional duplex systems without sacrificing the ability to detect and tolerate faults. The basis of our approach is virtual duplex systems. Virtual duplex systems use temporal rather than structural redundancy to detect faults. Experimental results show that such systems provide excellent fault detection in the case of transient failures. For permanent failures, the use of systematic diversity in combination with design diversity produces encouraging results. A virtual duplex system requires only a single processing node, whereas a traditional duplex system needs two. To match the performance of a traditional duplex system utilizing more than 50 percent of each processor’s performance, a virtual duplex system must use a faster processor. However, one faster processor is usually cheaper than two slower ones. In addition, a virtual duplex system requires only one instance of supporting circuitry (memory, I/O, and so forth) and uses a smaller printed circuit board. Because virtual duplex systems use temporal redundancy, their use might interfere with embedded applications’ real-time requirements. For example, consider an autonomous guided vehicle that uses a camera to control its movements: The embedded system controlling the vehicle must issue instructions at a rate proportional to the vehicle speed; that Jörg Keller Andreas Grävinghoff