The performance of a company's networked systems can have a significant effect on the productivity of the organization; however, typically few people on staff are qualified to manage such systems. As a result, it is important for systems managers to work as efficiently as possible. This dissertation investigates the use of Mobile Agents (MAs) as a means of improving network and systems management efficiency. The major objectives of this research are: (1) to simplify and improve the process by which a set of user-defined, MA-based management applications can be created and used, (2) develop original MA-based management applications to be compared against traditional stationary applications, and (3) investigate MA recovery protocols for responding to host or MA server failures.; To address the first major objective, a novel MA system called Resource Offering Agency and Mobility (ROAM) is designed and implemented. ROAM is a full-featured MA system supporting MA migration, inter-MA and inter-MA server communication, system security, MA recovery and a detailed user interface for managing local and remote ROAM servers and MA applications. ROAM also provides standard software classes which simplify the process of creating MA applications to perform typical distributed management tasks such as: data collection, monitoring and running tools. The design of ROAM also improves the process of creating MA applications by reducing the coding effort and avoiding potential programming concerns that occur with current MA systems with regards to the way security, inter-MA communication, MA migration and MA termination are implemented. Through examples and a discussion of the software architecture it is shown that ROAM provides an easy-to-use and expandable platform for creating MA applications.; To address the second major objective, MA-based management applications are developed. The primary MA application investigates using MAs to perform synthetic response time monitoring. In order to analyze this application, new mathematical models for estimating bandwidth usage were developed and monitoring experiments were run on academic and industry networks. The results of these analyses show MAs to be more efficient (lower bandwidth usage), flexible (behaviors can be quickly modified) and scalable (MAs can be quickly distributed to a new part of the network) compared with stationary monitors. The MAs also demonstrate their value as a means of signaling performance problems, aiding in debugging, and identifying performance trends for potentially every user in a local area network.; Four MA recovery protocols are analyzed to address the third major objective. For each protocol, new analytical models are developed for expected response time (the time from when an MA is launched until it completes its itinerary) and expected coverage (the number of hosts the agent is able to visit and complete its task). The results of this analysis are used to recommend recovery protocols for different types of MA application.
展开▼
