A Hybrid Fault Tolerance Method for Recovery Block with a Weak Acceptance Test

Abstract

Software reliability represents a major requirement for safety critical applications. Several fault tolerance methods have been proposed to improve software reliability. These methods are based on either fault masking such as N-version programming or on fault detection such as in the recovery block method. The success of the recovery block method depends on a high quality of the effective acceptance test, which is sometimes very difficult to achieve. In this paper, we propose a hybrid fault tolerance method called recovery block with backup voting to improve the reliability of the normal recovery block in the case of a weak acceptance test. In the proposed method, a copy of the outcome of each version is stored in a cache memory as backup, and when the recovery block method fails to produce a correct output due to a weak acceptance test, the stored values are used as inputs to a voting method to produce the correct output. A Monte Carlo based simulation method is used to show the reliability improvement in the new proposed hybrid method as well as to show the decreased dependency of the new method on the quality of the acceptance test, which makes the new method more suitable for critical applications where the construction of an effective acceptance test is difficult.