Abstract:

In order to improve the availability and quality of service （QoS） of existing cluster systems, a model of alternate service based on the system load is proposed by introducing a dynamic k-ray tree architecture. In this model, an auto-regressive load prediction model, a multi-machine heartbeat mechanism based on the election domain partition, and a leader election mechanism and algorithm in the dynamic domain are all employed. Theoretical analyses and experimental results demonstrate that the proposed model greatly improves the reliability and availability of cluster systems and saves the average response time cost and the election overhead. Thus, it well satisfies the requirements for high availability and QoS of large-scale Cluster services.

Key words: availability, quality of service, dynamic k-ray tree, election domain, alternate service