Optimal Agent Selection

The Internet contains a vast array of sources that provide identical or similar services. When an agent needs to solve a problem, it may split the problem into “subproblems” and find an agent to solve each of the subproblems. Later, it may combine the results of these subproblems to solve the original problem. In this case, the agent is faced with the task of determining to which agents to assign the subproblems.We call this the agent selection problem (ASP for short). Solving ASP is complex because it must take into account several different parameters. For instance, different agents might take different amounts of time to process a request. Different agents might provide varying “qualities” of answers. Network latencies associated with different agents might vary. In this paper, we first formalize the agent selection problem and show that it is NP-hard. We then propose a generic cost function that is general enough to take into account the costs of (i) network and server loads, (ii) source computations, and (iii) internal mediator costs. We then develop exact and heuristic based algorithms to solve the agent selection problem.