The reliable execution of mobile agents is an important design issue for building a mo-bile agent system and many fault-tolerant execution schemes have been proposed along roughly two approaches: replication and checkpointing. In this chapter, we proposed a new fault-tolerant execution model and analyzed its statistical properties. Our model ef-fectively combines available techniques and achieves better cost-effectiveness than existing models by eliminating redundant consumption of computational resources.
In our model, failures are classified into two classes: on-site failure and frontal failure.
This classification captures the intrinsic difference between these two classes of failures which causes different effects on mobile agents’ behaviors. An on-site failure is a node’s failure during an agent’s execution within it. It may block the agent and alter the agent’s state. On the other hand, a frontal failure is a node’s failure before an agent’s entrance to it which blocked the agent’s entry, but did not alter the agent’s state. For each kind of failures, an exceptional-event handling method is adopted. It can greatly decrease network resources consumption and improve the overall performance during mobile agent’s execution.
For the first time, the behaviors of mobile agents in networks that may contain faults are statistically analyzed in a quantitative way, which exploits a new approach to assess-ing the performance of agent-based systems. Several important parameters on system performance, including migration time, life expectancy, and population distribution of mobile agents, are analyzed. We also considered the performance of our model in reli-able networks. The analytical results reveal the theoretical insights into the fault-tolerant execution of mobile agents and show that our model can outperform the existing fault-tolerant models. The analytic method proposed in this chapter may also be applied for performance analysis in other complex systems. Our model is reliable and cost-efficient, which offers us a promising way for designing reliable mobile agent system.
Chapter 4
An Execution Prototype of Mobile Agent-Based Peer-to-Peer Networks
4.1 Introduction
Peer-to-peer networks are one of the trends in the field of internetworking. In p2p-networks, the hosts, or peers, act as client and server. Milojicic et al. [80] lists some characteristics of p2p-system. They are: decentralization; scalability; anonymity; self-organization; cost of ownership; ad-hoc connectivity; performance; security; transparency and usability; fault resilience; and interoperability.
Probably the most significant of these are Decentralization, Scalability and Ad-Hoc connectivity. As the clients act as servers in p2p-systems, there is no need for central management, which in traditional client-server model is done by servers. Scalability is achieved, as hosts can join or leave the network easily without having to register into a database. As hosts can also be up or down at every instant, ad-hoc connectivity is achieved.
Peer-to-peer networks can be divided into pure p2p-networks and hybrid networks.
In pure p2p-networks, there is no kind of central servers as in hybrid model where some servers are offered for e.g. locating the resources.
The most known p2p-systems are probably file-sharing networks, such as Napster, kazaa, DC++, Gnutella and Bittorrent. But p2p is also used in e-commerce, distributed computing and in instant messaging (such as MSN Messenger).
It can be seen that in a large communication network such as Internet, agents have to be generated frequently and dispatched to the network. Thus, they will certainly consume a certain amount of bandwidth of each link in the network. If there are too many agents migrating through one or several links at the same time, they will introduce too much transferring overhead to the links. Eventually, these links will be busy and indirectly block the network traffic. Therefore, there is a need of developing routing algorithms that consider about the traffic load. Since the state of different links may change dynamically over time, the agents have to dynamically adapt themselves to the environment, which increases the difficulty for both algorithm design and theoretical analysis. In [19], the network state is monitored by launching an agent at regular intervals from a source to a certain destination. In [22], the agent was enabled to estimate queuing delay without waiting inside data packet queues. In [71], the authors showed that the information needed in [19, 22] for each destination is difficult to obtain in real networks. In [9],
a mechanism of handling routing table entries at the neighbors of crashed routers was proposed which significantly improved the algorithm proposed in [19, 22]. In [16], the authors formulated a method of mobile agent planning, which is analogous to the travelling salesman problem [39] to decide the sequence of nodes to be visited by minimizing the total execution time until the desired information is found.
In this chapter, we propose an agent-based routing algorithm in which the traffic cost for each link is considered. To balance the traffic load on each link, we introduce the max-imum entropy theory into our algorithm to find an optimal probability distribution that makes inference on the known traffic information and balances the traffic load. Theoretical analysis shows that our derived probability distribution that an agent on an intermediate node may select a neighboring node and move to satisfies these two requirements. The remainder of this chapter is structured as follows. Section 4.2 introduce 4 most popular P2P systems. Section 4.3 presents the motivation of our research, i.e., why use mobile agents in peer-to-peer networks. Section 4.4 presents our algorithm. Section 4.5 intro-duces the maximum entropy theory. Section 4.6 provides theoretical analysis, Section 4.7 shows some simulation results, and Section 4.8 gives our conclusions.