自适应累加型失效检测模型.docVIP

自适应累加型失效检测模型研究 摘要:传统失效检测输出二值信息分别代表信任或怀疑,然而该机制存在灵活性不足的问题。针对这种问题,累加型失效检测以怀疑级别为输出,能适应同时运行的不同进程的qos需求。在分析和研究已有失效检测模型和累加型失效检测算法的基础上,提出一种新的累加型失效检测模型——exp-acc-fd。该模型利用幂律计算出心跳间隔的加权平均值,将该均值和距上次心跳到达的时间代入指数分布函数,从而计算出被监测进程的怀疑级别。实验分析表明,在相同的检测时间内,exp-acc-fd准确性高于nfd-e失效检测模型和phi失效检测模型。 关键词:失效检测;自适应;怀疑级别;指数分布;幂律 adaptive accrual failure detection model 英文作者名shi lei1*, chen wen-yuan1, tao yong-cai1, wei lin2 英文地址(1.school of information engineering, zhengzhou university, zhengzhou henan 450001, china; 2.school of software technology, zhengzhou university, zhengzhou henan 450002, china) abstract: the output binary information of traditional failure detection represents trust or suspicion respectively. however, the mechanism is lack of flexibility. for this problem, cumulative-type failure detection, taking the level of suspicion as output, can adapt to the qos requirements of different processes running simultaneously. an accrual failure detection model named exp-acc-fd was proposed on the basis of analyzing and researching the existing failure detection models and accrual failure detection algorithms. it calculates the weighted mean of heartbeat inter-arrivals with power law and substitutes the weighted mean and the time between last heartbeat coming and current time into exponential distribution to get suspicion level of monitored process. the simulation analyses show that the accuracy of exp-acc-fd is higher than nfd-e and phi within the same detection time. key words: failure detection; adaptive; suspicion level; exponential distribution; power law 0引言 失效检测指的是检测者利用失效检测模式及失效检测算法对被检测对象进行检测,从而判断被检测者是否失效的过程。分布式系统中,失效检测器是失效恢复、动态重启、可靠性通信、集群管理等功能的基础[1],在无线网络、云计算等领域也有重要的应用[2-4]。出现失效现象之后,需要有效的失效检测以及数据修复技术来延缓系统的平均数据丢失时间。然而,实际系统中很难在系统发生失效后立即进行数据修复,简单的做法是系统中的存储节点定期互相发送心跳信息以确定是否出现设备失效[5-6]。 传统的失效检测(如chen[7])输出一个二进制信息,用1和0分别代表信任和怀疑。在较为复杂的应用条件下,这种输出机制存在灵活性不足的问题[8]。针对这种缺点,défago等[9]提出一种新的失效检测模型,即累加型失效检测模型。这种失效检测对每一个进程都设置一个对应的值作为输出,这个值并不简单地判定信任或怀疑,而是代表对被检测进程的怀疑级别。监测进程可以根据不同的怀疑级别采取不同的措施,而不是等进程失效后再做