Implementation and performance analysis for emulation of causal ordering protocol on Internet

碩士 === 國立中正大學 === 通訊工程研究所 === 90 === In a distributed system, each site must communicate with other sites by exchanging messages in the synchronous way. Therefore, messages that are sent to and received by all sites must be ordered according to their cause-effect relation to ensure correc...

Full description

Bibliographic Details
Main Authors: wen-liang kuo, 郭文亮
Other Authors: Kim-Joan Chen
Format: Others
Language:zh-TW
Published: 2002
Online Access:http://ndltd.ncl.edu.tw/handle/82820294716362209211
Description
Summary:碩士 === 國立中正大學 === 通訊工程研究所 === 90 === In a distributed system, each site must communicate with other sites by exchanging messages in the synchronous way. Therefore, messages that are sent to and received by all sites must be ordered according to their cause-effect relation to ensure correct behavior of the system. However, due to unpredictable network delay, the ordering requirement of the messages may be violated at the receiving sites. To cope with the ordering issue, causal-ordering protocol was proposed. Moreover, in real-time distributed systems, not only all sites must maintain proper causal-ordering relation of messages but the life-time of the messages concerned as well. Here the -causal ordering protocol which is based on causal-ordering protocol addresses the real-time issue at the same time. Most research proposes new theories or new structures for particular applications. Few of them pay attention on the actual performance of these protocols on all sites in real-time distributed systems. In this paper, we studied related issues and implement the -causal ordering protocol on Internet and also on the emulated network. We have also observed the interaction of all sites under the protocol, verified the protocol behavior in different environments and research into the potential reason of observed phenomenon. Eventually, we have concluded our result about the choice of the  value. In this paper, we have found that the  value must be as possible as small, subject to the lowest throughput requirement of the real-time distributed system. If the  value is large, we have found that reliable retransmissions help some sites to decrease the overhead of synchronization. We believe our research will be beneficial to the improvement of the group communication in the future and can provide some helpful reference for the designer.