by Xu, Yijia.

Abstract (Summary)
Distributed simulations have been widely applied as the method to study complex systems which are analytically intractable and numerically prohibitive to evaluate. However it is not a trivia task to develop distributed simulations. Besides distributed simulations may introduce difficulties for analysis due to decentralized, heterogeneous data sources. It is important to integrate these data sources seamlessly for analysis. In applications for system design, it is required to explore the alternatives of hardware components, algorithms, and simulation models. How to enable these operations conveniently is critical for the distributed system as well. All these challenges raise the need of a workbench that facilitates rapid composition, evaluation, modification and validation of components in a distributed system. This dissertation proposes a platform for these challenges, which we refer to the SPEED-CS platform. The architecture of the platform consists of multiple layers that include network layer, component management layer, components layer, and modeling layer. It is a multi-agent system (MAS), containing static agents and mobile agents. The mobile agent is referred as the Data Exchange Agent, which is able to visit sub-simulations and has the intelligence to find the useful data for output analysis. Experiments show that the MAS requires much less network bandwidth than the “centralized” system does, in which simulations report data to output analyst. The application of the SPEED-CS platform is extended to handle systems with dynamic data sources. We demonstrate that the platform can be used for parallel 13 reality applications where simulation parameters can be updated according to realtime sensor information. Data exchange agents are involved to manage the collection, dissemination, and analysis of data from dynamic data sources including simulations and/or physical systems. The SPEED-CS platform is also implemented to integrate simulations and optimizations. The system is able to provide services to facilitate distributed computing, event services, naming services, and component management. One of the important features is that the component sets can be updated and enlarged with different models adding in. This feature enables the platform to work as a testbed to explore alternatives of system designs. Finally we conclude this dissertation with several future research topics. 14
Bibliographical Information:


School:The University of Arizona

School Location:USA - Arizona

Source Type:Master's Thesis



Date of Publication:

© 2009 All Rights Reserved.