Using agent-based software for scientific computing in the NetSolve system

Agent-based computing is increasingly regarded as an elegant and efficient way of providing access to computational resources. Several metacomputing research projects are using intelligent agents to manage a resource space and to map user computation to these resources in an optimal fashion. Such a project is NetSolve, developed at the University of Tennessee and Oak Ridge National Laboratory. NetSolve provides the user with a variety of interfaces that afford direct access to preinstalled, freely available numerical libraries. These libraries are embedded in computational servers. New numerical functionalities can be integrated easily into the servers by a specific framework. The NetSolve agent manages the coherency of the computational servers. It also uses predictions about the network and processor performances to assign user requests to the most suitable servers. This article reviews some of the basic concepts in agent-based design, discusses the NetSolve project and how its agent enhances flexibility and performance, and provides examples of other research efforts. Also discussed are future directions in agent-based computing in general and in NetSolve in particular.