An object-oriented approach for implementing algorithm-based fault tolerance

The authors demonstrate the practical use of an object-oriented system to incorporate fault tolerance and reliability into data objects. The object-based fault tolerance scheme uses abstraction to conceal algorithm-based fault tolerance layers. The scheme allows a layer of fault tolerance to be added to data objects without affecting or altering the use of the data objects. It is shown that the C++ class mechanisms of overloading and derivation permit the added fault tolerance to be transparent to the original data objects. To demonstrate the feasibility of this approach, using C++, a libray of matrix functions is presented and a layer of fault tolerance around matrix data objects is added. The weighted checksum code technique was implemented to create fault-tolerant matrix data objects. This allows programmers to add algorithm-based fault tolerance onto existing matrix applications without requiring modification to the original application. The implementation was experimentally evaluated using a software fault-injection tool, that emulated realistic hardware faults. An error coverage of over 96% was obtained with a memory overhead of 28%. The empirical results confirm the viability of the approach by demonstrating that object-based encapsulation is a valid method for transparently implementing algorithm-based fault tolerance.<<ETX>>