Browse Prior Art Database

Dual Redundancy with Checking

IP.com Disclosure Number: IPCOM000121904D
Original Publication Date: 1991-Oct-01
Included in the Prior Art Database: 2005-Apr-04
Document File: 3 page(s) / 73K

Publishing Venue

IBM

Related People

Abdelnour, GM: AUTHOR [+6]

Abstract

In machines where high availability is required, functions can be replicated such that failures can be tolerated. The usual method implemented is Triple Modular Redundancy (TMR). TMR provides three copies of the function. All copies are fed the same inputs. The three outputs are then fed into a voter. If two or more of the outputs are identical, the majority is chosen as the correct output.

This text was extracted from an ASCII text file.
This is the abbreviated version, containing approximately 66% of the total text.

Dual Redundancy with Checking

      In machines where high availability is required,
functions can be replicated such that failures can be tolerated.  The
usual method implemented is Triple Modular Redundancy (TMR). TMR
provides three copies of the function.  All copies are fed the same
inputs.  The three outputs are then fed into a voter.  If two or more
of the outputs are identical, the majority is chosen as the correct
output.

      The cost of TMR is more than three times that of a single copy.
What is needed is a method with nearly the same efficiency, in terms
of error tolerance and in speed, but at a lower cost.

      Dual Active Redundancy with Checking (DRC) provides two copies
of a function coupled with internal (to the function) error detection
mechanisms.  Examples of internal error detectors are:  parity, n out
of m codes, etc.

      Identical inputs are provided to both copies of a function.
The two outputs, including end error signal, are compared.  If they
agree, then one output is chosen, arbitrarily, as the true output of
the function.  If the outputs are different and the error signals are
also different, then the true output is chosen from the function
whose error output indicates that no error was detected.  If both
outputs indicate an error was detected, then the function has filed.

      From a reliability point of view, DRC is better than TMR.  Up
to 73% of a function's failure rate can be added to the function (to
a...