Summary: Chapter 1
Introduction
Fault­tolerant computing has traditionally been studied in the context of specific
technologies, architectures, and applications. One consequence of this tradition is
that several subdisciplines of fault­tolerant computing have emerged that are appar­
ently unrelated to each other: these subdisciplines deal with specific classes of faults,
employ distinct models and design methods, and have their own terminology and
classification [14, 40, 58]. As a result, the discipline itself appears to be fragmented.
Another consequence of this tradition is that verification of fault­tolerant
systems is often based on implementation­specific artifacts---such as stable storage,
timeouts, and shadow registers---without explicitly specifying what properties of
these artifacts are necessary. Such verification is imprecise and hence unsuitable,
especially for safety­critical systems.
Efforts have been made in the last decade to redress the problems described
above. Most of these efforts have focussed on uniformly classifying fault­tolerant sys­
tems, and two noteworthy classifications have emerged. One is based on a distinction
between the notions of faults, errors, and failures: faults in a physical domain can
cause errors in an information domain, whereas errors in an information domain can
1


Source: Arora, Anish - Department of Computer Science and Engineering, Ohio State University

Collections: Computer Technologies and Information Sciences