Topological theory of intelligent agent networks provides crucial information about the structure of agent distribution over a network. Agent network topologies not only take agent distribution into consideration, but also consider agent mobility and intelligence in a network. Research in the agent network topology area adopts topological theory from the distributed system and computing network fields, such as local area network (LAN) without considering mobility and intelligence aspects. Moreover, agent network topology theory is not systematic and relies on graph-based methodology, which is inefficient in describing large-scale agent networks.[1]
Definition
An agent network topology represents the information of agent distribution over an agent network, which incorporates agent mobility and intelligence aspects into the process of arranging and configuring an agent network.[2]
Related work
The term, agent network topology, is derived from mathematical topological graph theory. This concept overlaps with topological theory in data communications and distributed systems areas.
Existing topological theories in the computing field have been mainly applied to data communications and distributed systems areas. These theories have made some extraordinary contributions. However, as an emerging discipline, topological theory in multi-agent systems is inadequate. Existing topological theory cannot fulfill the needs of an agent network because an agent network has its specific characteristics, which include mobility, intelligence, and flexibility. Agent network topologies take not only agent distribution into consideration but also consider agent mobility in a network. Most of the current research work in the agent network topology area adopts topological theory from the distributed system and computing network fields without considering mobility and intelligence aspects.
Research work in the agent network topology area is also not systematic and relies on graph-based methodology. Graph-based topological analysis of a network topology is often based on the network graph provided and sometimes lacks precise measurements of each agent. Moreover, existing agent network topologies are incapable of providing detailed information about each agent and its relationship with other agents on a network. This increases the difficulty of agent communication and cooperation, such as agent searching or matching, over a network. Thus, the research direction of agent network topologies needs to follow these three characteristics based on some concrete network performance analysis.
Classification
Agent network topologies are primarily classified into two main categories, including simple agent network topology, and complex agent network topology.[3]
Simple
- Centralised agent network topology,
- Peer-to-peer agent network topology,
- Broadcasting agent network topology,
- Closed-loop agent network topology,
- Linear agent network topology,
- Hierarchical agent network topology,
- Heterarchical agent network topology.
Complex
- Regular network topology,
- Random network topology,
- Small-world network topology,
- Scale-free network topology.
In a regular network, nodes (agents) are distributed in order and the connections between nodes are based on certain constraints. For example, the wiring process is based on finding neighbour agents within the shortest distance. Regular network topology can describe simple networks but it is incapable of describing complex networks efficiently. Generally, regular network topology is limited to describe static networks.
Based on random graph theory, a random network topology can describe a large-scale complex network. It is more realistic than regular network in describing the real-world complex networks such as multi-agent systems. However, the limitation of random network topology theory is the difficulty of predicting, monitoring and controlling a network. For most agent-based systems this is unacceptable because most of the implemented agent-based systems require a high degree of monitoring and controlling. Therefore, Small-world network topology is suggested.
Small-world concept is becoming important in multi-agent systems, in which agents are often considered as nodes. It is difficult to use simple or regular agent network topology to describe an overall view of a large multi-agent system. Small-world topology could efficiently describe the conceptual view of a complex agent network. However, Small-world topology still lacks the ability to adapt to a dynamic environment. In other words, Small-world topology is not an ideal solution for the networks that are constantly changing.
The scale-free network topology, in which a network is allowed to change network connections dynamically and the nodes (agents) on the network are inhomogeneous.
References
- ^ Watts, D.J.; Strogatz, S.H. (1998). "Collective dynamics of 'small world' networks". Nature 393 (6684): 440–442. doi:10.1038/30918. PMID 9623998.
- ^ Zhang, H.L.; Leung, C.H.C.; Raikundalia, G.K. (2008). "Topological Analysis of AOCD-based Agent Networks and Experimental Results". Journal of Computer and System Sciences (Elsevier) 74 (2): 255–278. doi:10.1016/j.jcss.2007.04.006.
- ^ Classification of Intelligent Agent Network Topologies and a New Topological Description Language for Agent Networks (PDF). IFIP International Federation for Information Processing. 228/2007. Springer Boston. 2007. pp. 21–31. doi:10.1007/978-0-387-44641-7. ISBN 978-0-387-44639-4.