分布系统中多米诺效应的分析与消除

分布系统中发生故障后,错误信息在进程间的传播,以及由此引起的系统恢复过程中的多米诺效应是分布系统容错恢复技术所面临的最重要与最困难的问题.目前各种研究工作偏重于具体的系统动态恢复技术.本文的研究重点在于首次揭示多米诺效应本身固有的数量特征及其与分布系统技术参数间的联系,给出多米诺效应动态消除技术,引入系统额外开销的下限,并针对动态技术的缺点,探讨多米诺效应的静态消除技术,为解决困难开辟新途径.

ANALYSIS AND ELIMINATION OF DOMINO-EFFECT IN DISTRIBUTED SYSTEMS

The error propagation among processes and Domino-effect during the error recovery of a multi-process system are very important and difficult problems, which must be solved by the error recovery techniques in distributed systems. Most of the present researches in this field focus on how to eliminate Domino-effect dynamically. This paper reveals for the first time the inherent numerical properties of Domino-effect and their relation to the technical parameters of a distributed system. The possible minimum run-time overhead caused by dynamical elimination of Domino-effect is given. To avoid the run-time overhead, a static approach is presented, which provides a new possible way to overcome the difficulties.