Using multi-agent system for code and data propagation

• Main Author: Lupa, Aleksander
• Format: Others
• Language: English
• Published: Blekinge Tekniska Högskola, Avdelningen för programvarusystem 2008
• Subjects: multi-agent system; code and data propagation; agent migration; Computer Sciences; Datavetenskap (datalogi); Software Engineering; Programvaruteknik
• Online Access: http://urn.kb.se/resolve?urn=urn:nbn:se:bth-4410

This work presents the concept of code and data propagation in a multi-agent system. First, the concepts of agent and multi-agent system are defined and examples are presented. Also arguments for using agent approach are given and potential benefits are listed. Afterwards the idea of code and data propagation is defined and explained. Then some examples of real solutions are given along with propagation algorithms, which depict the way of introducing the concept into real system solutions. Afterwards the code and data propagation in a multi-agent system is described, which is in many cases based on the object migration. Discussion about this concept ends with describing the types of agent migration and giving some examples of systems with agent migration. Then three multi-agent environments are described and one is chosen to be the basis of the implemented application. Second part of work starts with description of systems principal objective, which is the distributed calculating of prime numbers. At the beginning, agents of the system are presented, and then the main system processes are depicted in detail. All algorithms are shown in sequence diagrams, which point all asynchronisms in the system. Afterwards migration phase is described with all migration types and algorithms. The experiments are conducted in two environments: home and university. The main aim is to find optimal configurations for both environments. The conclusion from this work is that introducing code and data propagation to a multiagent system in a form of agent migration in a heterogeneous network could considerably decrease the execution time. Moreover based on the efficiency vector of computers participating in the experiment there is a possibility to set a task distribution, which is close to optimal, without searching for optimal configuration every time when running the experiment.