Analysis and Evaluation of a New Algorithm Based Fault Tolerance for Computing Systems

Analysis and Evaluation of a New Algorithm Based Fault Tolerance for Computing Systems

Hodjat Hamidi (University of Isfahan, Iran), Abbas Vafaei (University of Isfahan, Iran) and Seyed Amir Hassan Monadjemi (University of Isfahan, Iran)
Copyright: © 2012 |Pages: 15
DOI: 10.4018/jghpc.2012010103
OnDemand PDF Download:
$30.00
List Price: $37.50

Abstract

In this paper, the authors present a new approach to algorithm based fault tolerance (ABFT) for High Performance computing system. The Algorithm Based Fault Tolerance approach transforms a system that does not tolerate a specific type of fault, called the fault-intolerant system, to a system that provides a specific level of fault tolerance, namely recovery. The ABFT techniques that detect errors rely on the comparison of parity values computed in two ways, the parallel processing of input parity values produce output parity values comparable with parity values regenerated from the original processed outputs, can apply convolution codes for the redundancy. This method is a new approach to concurrent error correction in fault-tolerant computing systems. This paper proposes a novel computing paradigm to provide fault tolerance for numerical algorithms. The authors also present, implement, and evaluate early detection in ABFT.
Article Preview

Introduction

Algorithm based fault tolerance (ABFT), proposed by Huang and Abraham (1984), is a fault tolerance scheme that uses Concurrent Error Detection (techniques at a functional level). System level applications of ABFT techniques have also been investigated (Acree, Ullah, Karia, Rahmeh, & Abraham, 1993; Banerjee, Rahmeh, Stunkel, Nair, Roy, & Abraham, 1990). These techniques assume a general fault model which allows any single module in the system to be faulty (Huang & Abraham, 1984). ABFT is widely applicable and it has proved its cost-effectiveness especially when applied to array processor (Jou & Abraham, 1986). Fault detection and diagnosis are integral parts of any fault tolerance scheme. There are two ways to detect faults:

  • (1)

    By off-line checking.

  • (2)

    By concurrent checking.

In an off-line checking scheme, the process is checked for its correctness while it is not performing any useful computation. This approach has the beneficial that the performance of the computer will be unaffected by the checking operation; however, this kind of checking can detect only permanent faults. Transient faults, which constitute 76-81% of faults in a Digital system (Jou & Abraham, 1986), will not be detected by off-line checks. So that detects transient faults, concurrent error detection schemes such as duplication and comparisons have been suggested. It has been observed that ABFT techniques are very cost effective when applied to processes elements. It may be noted, however, that the application of ABFT techniques is not limited to multi-process systems; they are also applicable to algorithms running on single processes, probably with less efficiency. An algorithm executing on a multiple process system is specified as a sequence of operations performed on a set of processes in some discrete time steps.

Complete Article List

Search this Journal:
Reset
Open Access Articles: Forthcoming
Volume 9: 4 Issues (2017)
Volume 8: 4 Issues (2016)
Volume 7: 4 Issues (2015)
Volume 6: 4 Issues (2014)
Volume 5: 4 Issues (2013)
Volume 4: 4 Issues (2012)
Volume 3: 4 Issues (2011)
Volume 2: 4 Issues (2010)
Volume 1: 4 Issues (2009)
View Complete Journal Contents Listing