Dynamic Adjustment of Dependability and Efficiency in Fault-Tolerant Software

In this paper we discuss the problem of attaining a dynamic compromise between using redundancy to improve software dependability and limiting the amount of redundancy so as to avoid unnecessary inefficiencies. A scheme, called self-configuring optimal programming (SCOP), is developed. SCOP attempts to reduce the resource cost of faulttolerant software, both in space and time, by providing designers with a flexible redundancy architecture in which dependability and efficiency can be adjusted dynamically at run time. A design methodology is proposed to introduce support techniques for such dynamic adjustment. Our scheme also suggests a general control framework into which design diversity, data diversity and multiple copies can be incorporated selectively. A detailed dependability and efficiency evaluation shows that SCOP can achieve the same dependability level as those of other existing schemes, while making more efficient use of available resources.