A multiplecopy data storage algorithm is presented that is designed specifically for reliability in a broadcast environment. The algorithm exploits redundancy and is meant for systems in which duplication cost is ou...
详细信息
A multiplecopy data storage algorithm is presented that is designed specifically for reliability in a broadcast environment. The algorithm exploits redundancy and is meant for systems in which duplication cost is outweighed by the need for added reliability. The purpose of the algorithm is not to provide transaction concurrency control, but to handle the problem of the mutual consistency of the copies. It is a loosely coupled algorithm in that it is designed to function even though not all copies of the data need agree and not all processors may be functioning properly. Voting is used to distinguish outdated or erroneous copies from correct ones. An important aspect of the proposed algorithm is that if no errors have occurred, the redundancy is largely transparent to the procedures that access the data, requiring no additional communication. Extra messages are needed when errors are detected in copies of stored information; this is done by combining the null message technique with the broadcast technology. A Markov analysis is performed to model the system's failure behavior.
暂无评论