Impact of hardware and software faults on ARQ schemes-an experimental study

Abstract

An understanding of the impact of hardware and software faults (HSFs) on any application is critical to the design of efficient software fault tolerance techniques. This paper considers HSFs and, by using a fault injection experiment, studies their impact on an automatic repeat request (ARQ) scheme in terms of throughput degradation. An ARQ scheme is used for error control in computer networks and is implemented in the data link layer. Our study shows that even in the absence of errors in the communication channel the throughput may be degraded with HSFs. Furthermore, we need to identify certain critical variables as locations for fault injection thereby elevating the fault models to a higher level of abstraction. The variables lie in the active path of the program and help accelerate the failure process. This results in fewer runs being needed in conducting the fault injection experiment. Since the accelerated failure process represents the worst case scenario for the fault models considered, the experiences would enable the fault tolerance engineer in designing/choosing fault tolerance mechanisms.