具容錯能力之網路管理系統設計

Abstract

網路管理系統之任務是提供快速且有效率的服務，以確保網路通訊品質，協助網路管理者監控整個網路系統之運作，為提升網路管理系統之「自動化」、「即時性」、「耐衝擊性」與「降低系統複雜度」的需求，有人提出利用行動代理人於網路管理系統之構想。然而現今之研究成果大都偏重於行動代理人行動力之應用，對代理人的學習能力之特性則較少探討，許嘉容學姊因而提出『互助學習式網路管理系統』之構想，可大幅提升網路管理系統之自動化以及學習與合作的能力，然而應用其構想於障礙管理尚有兩個主要問題未解決：(1) 障礙排除流程自動化以及障礙管理知識庫之設計與部署，及(2)行動代理人以及網路與管理節點之容錯機制﹔前者影響解決問題之速率，後者則防止無代理人或是代理人無法執行任務之情況發生，本研究旨在解決此二問題。 本研究首先依據障礙解決流程，規劃行動代理人、區域管理者及系統管理者在障礙排除過程之分工、合作及學習機制，進而由障礙排除效率及系統容錯角度規劃障礙排除知識在行動代理人、區域管理者及主管理者之分佈，並提出利用代理人互相備份合作的機制，以使障礙管理具容錯性質。 本研究所提出之障礙管理機制及系統容錯機制已在區域網路系統上驗證其可行性及實用性，且證實應用行動代理人之自動學習及合作機制確可大幅提升障礙排除效率及降低系統容錯機制設計之複雜度。

Network Management System (NMS) should provide some mechanisms to help network managers to monitoring and control the operation of a communication network system so that it will provide users effective, efficient, and reliable communication services . In order to enhance the capabilities of “automation”, ”real time”, robustness”, and “low system complexity” of the network management system, some researchers proposed to using mobile agent technique to implement network management system. However, most researches paid more attention on the mobility capability of mobile agents rather than the learning, and cooperation capability. Recently, Miss Hsu [15] proposed 『Collaboratively Learning Network Management System』which use the learning capability of the mobile agent to enhance the intelligence of the network management system. But there are two shortages in this system: (1) the fault management system and fault knowledge base are not designed, and (2) fault tolerance issue of network management system is not discussed. The purpose of this research is to solve these two problems. Based on the fault recovery procedures, the cooperation and learning mechanism of mobile agent, local manager and system manager of fault management system is designed first, Then the deployment of knowledge base on mobile agent, local manager, and main manager is designed. Finally, corporation capability of mobile agents and backup mechanism of fault knowledge base are used to implement the fault tolerant mechanism of the fault management system. The proposed network Fault management system has been implemented in a local area network to prove its feasibility. From this study, we find that the learning and cooperation capabilities of mobile agents can largely reduce fault recovery time and reduce the design complexity of fault tolerant network management system.