Autori: Fl. Popentiu - Vladicescu, P. Sens, P. Thyreod & G. Albeanu

Editorial: E. Zio, M. Demichela, N. Piccinin, Proceedings of the European Conference On Safety and Reliability, ESREL 2001, p.1031 - 103, 2001.

Rezumat:

The problem of software architecture according to the software reliability forecasting is considered. The proposed technique is based on a chain of automatic data collection, which allows us the possibility to adjust, during the execution, the strategy of fault management. Four modules are chained: monitoring, statistical, prediction, selection. The statistical module deals both with artificial neural networks (NN) and explanatory variables approaches (EV). The predictions are used to recalibrate/training the initial model. The selection process identifies the appropriate algorithm for adaptability of fault management. This paper addresses the structure of the statistical approach module from the viewpoint of the „predictions” module. Te main advantage of the proposed approach is that it gives the possibility to include, in the prediction process, information concerning the structure and the history of the distributed system behavior.

Cuvinte cheie: Distributed Systems, Neural Networks, Explanatory Variables, Statistical Estimation