Research And Implementation On Disaster-recovery Oriented Failure Detection Algorithm

With the rapid development of the information system and network technology, information system and network have become fundamental building blocks to accomplish daily jobs. When disaster from nature or made by people attacks centralized data and network, the consequence frequently is too far gone. Hence, lots of people pay more and more attention on the methods to prevent attack from disaster and prolong the runtime of operation.Failure detection is one of key technologies of the disaster recovery system, fast, efficient and accurate failure detection is the precondition and guarantee of the realization of the disaster recovery effectively. This paper researches the failure detection technology orienting disaster recovery working in WAN. First of all, it analyses several typical models of the distributed system, failure models of the distributed system, failure detection models of distributed system and the evaluation criterion of the failure detection algorithm. Then it analyses failure detection algorithms in the field of high availability and grid, and the knowledge in the field of large-scale condition.The main work of this paper is as follows:First of all, We Analyze detailedly a failure detection model orienting disaster recovery, and bring forward some problems needing to solve in this model. In order to solve these problems, this paper designs a new failure detection algorithm named HB-DR(Heartbeat For Disaster Recovery), which detects each other at regular intervals, and uses node and network's status to output a quantitative value stands for serviceability of disaster recovery system, and uses comparison between quantitative output and threshold value to estimate a disaster recovery system's serviceability. This algorithm could detect system's dependability in the disaster recovery model, which has been tested in analogue environment and has been proved that HB-DR could solve the problem of the debasement of accuracy caused by delay or lost of the messages.Secondly, On the basis of demand disaster recovery system asks for in large-scale and the characteristics of HB-DR algorithm, this paper makes a new design in large-scale environment, which uses neighbor model and notice between groups to improve the usability of system and reduce the workload.At last, A complete failure detection analogue environment is realized through Linux-HA project integrated HB-DR. We implement the detection and management of the resource in prototype system, and test the algorithm.