Detectores de Defeitos Não Confiáveis
Résumé
The consensus protocol, widely used to the coordination of fault-tolerant systems, have a big restriction when used in asynchronous environments. The impossibility of distributed consensus in asynchronous systems shown in 1985, proved that the consensus can not be solved in a deterministic way in such environments, due to the impossibility of distinguishing faulty processes from slow ones. The failure detectors represent a way to circumvent this problem, and even being inexact, they augment the information's amount on the system, assisting the accomplishment of the consensus and other protocols with the same problem. This work reviews the literature on detectors, presenting their principles and some proposals to their implementation, under different models of faults in the system. Moreover, this work makes an evaluation on the practical aspects from the detectors mainly ignored by the bibliography, as for example, the detection's granularity and the interaction with other protocols. A prototype of one of the studied proposals was also implemented, using the Java language, objectifing to verify its functional behavior. We also envisage to evaluate which are the main points to be improved, to increase its performance and transparency of usage.
Loading...