Cluster Operating System High Availability Services Research

With the scale expanded and the node number increased, the integral reliability of cluster system decreases correspondingly, and therefore, failure of nodes is inevitable. Meanwhile, the cluster system availability is more demanded with the cluster application getting more popular and especially with the fast development in the commercial application services. As the most foundamental system software in the cluster system, cluster operating system is built based on the node operating system and provides the interface for users to access the cluster services. In all, the cluster operating system needs to provide the HA services, and the good scalability for the HA service is a necessity due to the cluster scale growing.The "Dawning 4000" Cluster Operating System is a unified service-based one established on component technology, and mainly aims at high availability and good scalability in design. Interaction not only among components but also between system and users are based on service-access, so as to achieve the location-unaware communication. The purpose of this paper is to achieve the high availability services for "Dawning 4000". By analyzing the characteristics of cluster system and introducing the HA theories, a hierarchical architecture, with the features both distributed and centralized, is proposed. Then the group service component, which is the kernel component for the HA services, is designed and implemented. This design solves not only the problem of availability in cluster OS but also the scalability problem of HA services. The "Dawning 4000" cluster OS is practically deployed on 640 nodes.The research background and purpose are described in this paper. Then the HA theories, HA solution, and the technologies and key issues to realize cluster HA are introduced. Focusing on the key issues of the HA services in the cluster established on the component technology, the major part of this paper is arranged as follows. The paper introduces the background and objective of this project, and then, introduces fundamental HA theories needed in this paper, the solutions, technologies and key issues to realize HA for the system. Focused on the key issues for the componentationized cluster OS to provide HA services, this paper carefully describes the approaches in doing this, the functions and importance of group-service component, and also, the design and implementation of such component. Finally, quantitative analysis is done to the HA services of the cluster OS based on mathematical model.