Chinese Journal of Computers   Full Text
Title	High-Availability Model Based-on SonD and Its Availability Allocation Algorithms
Authors	TANG Hai-Ying1),2) XU Lu1)
Address	1)(Institute of Computing Technology, Chinese Academy of Sciences, Beijing 100080) 2)(Graduate University of Chinese Academy of Sciences, Beijing 100039)
Year	2007
Issue	No.10(1731—1739)
Abstract &   Background	Abstract Limited scalability is a common problem for traditional high availability systems. With management techniques of Blue Whale SonD system, a High-Availability model based-on SonD is proposed in this paper. This model has better scalability for system management and resource utilization comparing with traditional high availability system models. With the dynamic mapping mechanism between computing resources and storage resources provided by Blue Whale SonD system, this model implements a kind of virtual high availability server by separated management of computing resources and storage resources for various services, releasing the complexity of management and deployment. Since redundant resources could be shared by different services, this model also improves the resource utilization ratio. The key issue of High-Availability system based-on SonD is how to allocate server resources for various services to satisfy the different availability requirements and runtime modes of each service. This availability allocation problem is different from traditional redundancy allocation problems on allocation pattern. According to the specific availability allocation pattern of this system, a Best-Fit Redundancy-Prior allocation algorithm is proposed, which aims to find better solution when considering values of both cost and penalty in objective function. The experiments show that this algorithm achieves the allocation requirement of real time service deployment practically. keywords SonD system; high availability system; availability evaluation; resource utilization evaluation; availability allocation background Most current high availability systems use direct attached storage model for their storage sharing. The scalability of these systems is limited by this storage model. The availability management and deployment of large scale system would be complex. The resource utilization ratio also would be low due to the separated redundant resources of each sub-system. Blue Whale SonD system uses network storage model, and implements dynamic mapping mechanism between computing resources and storage resources. These techniques enable SonD system to provide a scalable infrastructure for high availability system. This paper proposes a high availability model based on SonD system. In this model, virtual high availability servers, composed of computing resources and storage resources, are allocated to various services. The availability of basic computing resources and storage resources is maintained by management servers of SonD system. The basic redundant resources could be shared by some services. How to allocate the resources to various services to meet their requirement on availability, while increase the resource utilization ratio at the same time, is key to availability deployment system, especially for those with large scale resources. The optimal redundancy allocation problem, or optimal reliability allocation problem, has been proved to be a NP-hard problem. Many heuristic algorithms have been developed to solve this problem. This paper works on the availability allocation problem, which is different from traditional allocation problems in its system structure and allocation patterns. Especially, one important function of the authors’ system is to implement real-time deployment of various services. Thus, the runtime of the allocation algorithm is more important than how close the final solution will be to the optimal solution. A good solution which could be obtained in a short time is better than an optimal solution which will be obtained after a long time. A Best-Fit Redundancy-Prior allocation algorithm is proposed to solve the availability allocation problem. The experiments show that this algorithm finds a good initial solution for availability allocation problem This work is partially supported by the National Basic Research Program (973 Program) of China under grant No.2004CB318205. This project involves research on service-on-demand deployment model and service quality of network storage. The work in this paper is part of the research on service-on-demand deployment model of network storage, focusing on the reliability and availability of service deployment.