By Topic

An algorithm-based error detection scheme for the multigrid method

Sign In

Cookies must be enabled to login.After enabling cookies , please use refresh or reload or ctrl+f5 on the browser for the login options.

Formats Non-Member Member
$31 $13
Learn how you can qualify for the best price for this item!
Become an IEEE Member or Subscribe to
IEEE Xplore for exclusive pricing!
close button

puzzle piece

IEEE membership options for an individual and IEEE Xplore subscriptions for an organization offer the most affordable access to essential journal articles, conference papers, standards, eBooks, and eLearning courses.

Learn more about:

IEEE membership

IEEE Xplore subscriptions

2 Author(s)
Mishra, A. ; Bradley Dept. of Electr. & Comput. Eng., Virginia Polytech. Inst. & State Univ., Blacksburg, VA, USA ; Banerjee, P.

Algorithm-based fault tolerance (ABFT) is a technique to provide system level error detection and correction on array processors as well as multiprocessors at a low cost. Since the early 1980s the technique has been extensively applied to several linear algebraic algorithms, e.g., matrix multiplication, Gaussian elimination, QR factorization, and singular value decompositions, etc. An important class of problems in numerical linear algebra dealing with the iterative solution of linear algebraic equations arising due to the finite difference discretization or the finite element discretization of a partial differential equation, however, has been overlooked. The only exception is the recent application of algorithm based error detection (ABED) encodings to the successive overrelaxation algorithm for Laplace's equation. In this paper, ABED is applied to a multigrid algorithm for the iterative solution of a Poisson equation in two dimensions. Invariants are created to implement checking in the relaxation, the restriction, and the interpolation operators. Modifications to invariants due to roundoff errors accumulated within the operators, which often lead to a situation known as false alarms, have been addressed by deriving the expressions for the roundoff errors in the algebraic processes in the operators and correcting the invariants accordingly. The ABED encoded multigrid algorithm is shown to be insensitive to the size and the range of the input data besides providing excellent error coverage at a low latency for floating-point, integer, and memory errors.

Published in:

Computers, IEEE Transactions on  (Volume:52 ,  Issue: 9 )