A decentralized multi-agent based network management system for ICTD networks

Abstract

Network management is fundamental for assuring high quality services required by each user for the effective utilization of network resources. In this research, we propose the use of a decentralized, flexible and scalable Multi-Agent based system to monitor and manage rural broadband networks adaptively and efficiently. This mechanism is not novel as it has been used for high-speed, large-scale and distributed networks. This research investigates how software agents could collaborate in the process of managing rural broadband networks and developing an autonomous decentralized network management mechanism. In rural networks, network management is a challenging task because of lack of a reliable power supply, greater geographical distances, topographical barriers, and lack of technical support as well as computer repair facilities. This renders the network monitoring function complex and difficult. Since software agents are goal-driven, this research aims at developing a distributed management system that efficiently diagnoses errors on a given network and autonomously invokes effective changes to the network based on the goals defined on system agents. To make this possible, the Siyakhula Living Lab network was used as the research case study and existing network management system was reviewed and used as the basis for the proposed network management system. The proposed network management system uses JADE framework, Hyperic-Sigar API, Java networking programming and JESS scripting language to implement reasoning software agents. JADE and Java were used to develop the system agents with FIPA specifications. Hyperic-Sigar was used to collect the device information, Jpcap was used for collecting device network information and JESS for developing a rule engine for agents to reason about the device and network state. Even though the system is developed with Siyakhula Living Lab considerations, technically it can be used in any small-medium network because it is adaptable and scalable to various network infrastructure requirements. The proposed system consists of two types of agents, the MasterAgent and the NodeAgent. The MasterAgent resides on the device that has the agent platform and NodeAgent resides on devices connected to the network. The MasterAgent provides the network administrator with graphical and web user interfaces so that they can view network analysis and statistics. The agent platform provides agents with the executing environment and every agent, when started, is added to this platform. This system is platform independent as it has been tested on Linux, Mac and Windows platforms. The implemented system has been found to provide a suitable network management function to rural broadband networks that is: scalable in that more node agents can be added to the system to accommodate more devices in the network; autonomous in the ability to reason and execute actions based on the defined rules; fault-tolerant through being designed as a decentralized platform thereby reducing the Single Point of Failure (SPOF) in the system.