Abstract:
Nodes in a Distributed system (DS) are prone to failure due to link failure, resource failure
or any other reason have to be tolerated for the execution of the system smoothly and
accurately. Faults may affect the operation of DS at any time. In this regard diagnosing the
faulty nodes in the distributed system is one of the requirements to make the system more
reliable and efficient. Faults can be detected and recovered by many techniques according
to the requirement.
In this thesis we presented a robust stochastic search method based on evolutionary
algorithmic approach to detect faults in a k-connected DS with at most (n-1)/2 permanent
faulty components out of ‘n’ tested components using two approaches. The first approach
is by using a Genetic algorithm with Fuzzy controlled mutation and Tournament selection
which provides how a fuzzy-logic can be applied to the algorithm in order to detect
permanent faults in a k-connected distributed System. The second approach is a parallel
evolutionary algorithm based on immunity theory, which provides how parallel evolution
is efficient in this context.
The program is implemented in C++ and was tested with random test graphs. Empirical
results reveal that the proposed immune parallel evolutionary algorithm (IPEA) method
for fault detection in DS is more efficient when compared to fuzzy genetic algorithm
(FGA) and the standard basic genetic algorithm (BGA) approaches. The results are
satisfactory and demonstrated the efficiency of genetic operators combined with parallel
evolution and immune selection used for this study than FGA and BGA.