Conception et analyse de schemas de redondance pour grappe d'ordinateurs (French text).

摘要

Most of the search undertaken in the field of the clusters of computers are intended for the scientific applications, that is to say for the high performance. The theories on the availability of clusters are rare, if not non-existent, because in the scientific field the profits of the continuity of service are less obvious. Through this research, our objective is to set up analytical models allowing us to implement and analyze redundancy schemes ensuring the continuity of service of the cluster, and their impacts on the overall performance of the cluster. The key word in clusters for high availability is the redundancy. Associated with the concept of failover, it would allow to reach availability of few seconds of loss of service per year. In this thesis, we focus ourselves on the redundancy of the components allowing the processing of the requests (processors), and the data redundancy in order to improve the availability of the cluster. (Abstract shortened by UMI.)