As computer networks grow in scale, scope, and importance and as their topologies become more dynamic, the complexity of configuring, administering, and utilizing networks grows as well. We are studying new ways of thinking about and designing networks and network applications; in particular, we are building systems of cooperating mobile software agents. Mobile agents are programs that are free to move themselves and their thread of execution from computer to computer [20]; the environment the agents live in is the computer network. Agents can also interact with each other, sharing information to better accomplish their goals.
We believe that mobile agents offer exciting possibilities for the design of next-generation systems. The modularity of the mobile agent approach encourages the design of flexible, adaptive architectures. The malleability and dynamic reconfigurability of movable code allow us to re-tailor systems in the real world by adding to and changing running systems. And the metaphor itself -- mobile, encapsulated programs -- provides a new framework for thinking about computer networks.
This paper investigates a straightforward problem for a system of cooperating agents. We present results from an initial round of simulations along with a discussion of some strengths and weaknesses of this particular approach. The problem, building a map of a computer network, was chosen because it servers as a good test case for studying the behavior of mobile agents working together to solve a problem.
The mapping of network topology is a problem suitable for exploring the dynamics of mobile, multi-agent systems. The data of interest, the connectivity of nodes in the network, is inherently distributed and decentralized. The potential users of that data (for example, an application that needs to open a socket between a specific pair of hosts) are likewise spread across the system. Typically, routing problems are solved by collecting the far-flung data of network connectivity in a centralized store. Our experimental alternative is to send mobile agents out into the network to discover information about connectivity, building a map which can be used to make routing decisions. This decentralized solution presents an interesting alternative for network mapping and routing, particularly in cases where traditional centralized solutions are difficult to implement because the network topology is itself dynamic.
