Authors : M. Aliouat , M. Nekkache and Z. Aliouat
Abstract: Designing fault tolerant distributed systems, particularly when taking into account both hardware and software faults, is a challenge requiring extra endeavor. In this effect, backward error recovery approach is enough general to be used to recover a failing distributed application. In this study, we have to deal with the fastidious problem of interprocess rollbacks propagation during recovery operation. The control of this undesirable situation, so called domino effect problem, with its qualitative and quantitative aspects, has been well addressed in the past and has led to many proposal schemes for its total prevention or minimization. Among these schemes, we focus attention to Randell�s conversation and Kim�s Programmer Transparent Coordination of recovering concurrent processes (PTC). In this study, we present a variant of PTC approach aiming to make recovery operation more trustworthy and efficient. This improvement is reached towards new concepts namely: S-propagation, accusation limit, definite invocation and 2 by 2 reliability. The definite invocation and 2 by 2 reliability are used to avoid the domino-effect qualitative aspect, while the Accusation limit should prevent the domino effect quantitative aspect and should ensure also that the S-propagation will be stopped at a satisfactory reliability level. Therefore, we deduce a parameter noted Accusation limit Level which controls the maximum number of accused recovery regions per any process involved in rollback. Our variant is as flexible and efficient as PTC but more reliable. Furthermore, the computation validity obtained through the acceptance tests, is ensured by a number of tests greater than in the conversation scheme.
M. Aliouat , M. Nekkache and Z. Aliouat , 2006. An Efficient and Reliable Parameterizable Recovery Approach for Distributed Applications. International Journal of Soft Computing, 1: 247-254.