Although there are many projects focusing on multiagent systems, there are only a few focusing on systematic design of large scale multiagent system. In this paper we formalize the knowledge representation and sharing of agents, using symbol structures, define agencies as organizations (i. e. , a coalition of agents), propose a formalism to represent organizational Intelligence, devise a basic configuration for generalized agents (AG), and use them in a large scale multiagent system design. Private knowledge of an AG agent is represented by a symbol structure (SS) and AG agents can share their knowledge using combination, specialization and generalization methods that operate on the SS. Opposite to the other works, organizational knowledge, is defined as a property of at least a pair of AG agents.

This paper discusses the characteristics of human design knowledge. By studying a number of actual human made designs of excellent designers, the most frequent basic mental operations of a typical human designer have been found. They are: a design rule for hierarchical detailing reported previously, a micro design rule for generating a hierarchical expansion, dictionary operations to build a micro design rule and dictionaries. This study assumes a multiplicity of knowledge based on Zipf's theory, "the principle of least effort. " Zipf's principle may be proved and it becomes possible to understand the fundamental nature of human design.

This paper discusses a framework for automating fault management using distributed software agents. The management function is distributed among multiple agents that can carry out advanced reasoning activities on the network domain. Network domain modeling using Bayesian network is introduced. The agent detects, correlates and selectively seeks to derive a clear explanation of the alarms generated in its domain. Depending on the network's degree of automation, the agent can even carry out local recovery actions. The ideas of the paper are implemented in a software for inference in Bayesian network. We identify the potentialities of learning in the agent model, and present the class of problems to be addressed.

A Multi-Agent Learning Language (MALL) is defined as being necessary for agents in environments where they encounter crucial situations in which they have to learn about the environment, other parties moves and strategies, and then construct an optimal plan. The language is based on two major factors, the level of certainty in fully monitoring (surveying) the agents and the environment, and optimal plan construction, in an autonomous way. Most of the work related to software agents is based on the assumption that other agents are trustworthy. In the growing Internet environment this may not be true. The proposed new learning language allows agents to learn about the environment and the strategies of their opponents while devising their own plans. The language is being tested in our project of software agents for Electronic Commerce that operates in various security zones. The language is flexible and adaptable to a variety of agents applications.

Software agents are knowledgeable, autonomous, situated and interactive software entities. Agents' interactions are of special importance when a group of agents interact with each other to solve a problem that is beyond the capability and knowledge of each individual. Efficiency, performance and overall quality of the multi-agent applications depend mainly on how the agents interact with each other effectively. In this paper, we suggest an agent model by which we can clearly distinguish different agent's interaction scenarios. The model has five attributes: goal, control, interface, identity and knowledge base. Using the model, we analyze and describe possible scenarios; devise the appropriate reasoning and decision making techniques for each scenario; and build a library of reasoning and decision making modules that can be used readily in the design and implementation of multiagent systems.

Server performance is a major issue in improving the overall performance of the World Wide Web (WWW). This article introduces a dynamic mirroring-based approach to improve WWW servers' performance. In contrast to static mirroring, where mirror servers are allocated statically, our mirror servers' setup is driven by network traffic measurement. Performance in terms of latency is inferred from a queuing model. According to this model we show that latency of an overloaded server can be tuned by delegating a portion of the load to a cooperative mirror server. Cost is evaluated by the amount of load hosted by the mirror servers. The goal is then to keep the latency within a tolerable threshold, while minimizing the delegated load. This problem is formulated as a constrained optimization problem where the task is to assign a portion of load corresponding to each document to each mirror server. As the result of this work, we will have a balanced load among the servers, and a smoother traffic along the Internet, as well. Empirical results show that this approach can guarantee to maintain the performance while showing a significant decrease in the amount of load transferred to the mirror servers.

A major topic in the field of network and telecommunications is doing business on the World Wide Web (WWW), which is called Electronic Commerce (EC). Another major topic is blending Artificial Intelligence (AL) techniques with the WWW. In the Ex-W-Pert Project we have proposed an agent model for EC components that blends the traditional expert systems' reasoning engine with a multi-layer knowledge base, communication and documentation engines. In this project, EC is viewed as a society of software agents, such as customer, search, catalog, manufacturer, dealer, delivery and banker agents, interacting and negotiating with each other. Each agent has a knowledge-base and a reasoning engine, a communication engine and a documentation engine. The knowledge-base is organized in three layers: skill layer, rule layer and knowledge layer (S-R-K layers). In this project, for each EC agent, we identify the class of problems to be solved and build the knowledge base gradually for each layer. We believe that using this multi-layer knowledge base system will speed up the reasoning and ultimately reduce the operation costs.

Network traffic characteristics impacts directly network performance, and resource allocation policies. In this work, we introduce a multi-agent system, that manages the performance of web servers with minimal cost of mirroring. In our proposed system each web server is viewed as a software agent that perceives its environment by monitoring its traffic. The goal of the agent is to manage the performance, using cooperative mirror servers, while minimizing the cost of mirroring. Communication between the agents enables each web server to decide about its future actions, which is whether to share its load with the cooperative mirror servers, and how much load to assign to them. The architecture of a software agent that is intended to manage the performance of a web server, is elaborated and its different modules are described. Also a set of cooperative agents is defined, that form a multi-agent system and is intended to assure maintaining the performance with minimal cost of mirroring. The experimental results presented in this article illustrates the effectiveness of the proposed system.

This paper proposes a framework for distributed network management by incorporating fault and performance management metrics in a hierarchical decision making model. The goal of this research is to automate the fault management process. The fault management system is organized as a three level information processing model. Correlation results from each level are provided as evidence to the next level. Causal and temporal relationships between monitored variables are captured using Dynamic Bayesian Networks. As evidence is gathered, the probability of the presence of a fault is either strengthened or weakened. The proposed model is used for proactive fault detection as well as fault isolation purposes. A prototype implementing the ideas is presented.