Query restart strategies for Web agents

With the explosive growth of the internet, autonomous agents will increasingly need strategies for efficiently retrieving information. The time an agent (or server) takes to answer a query issued to it is often a random variable, whose distribution can be estimated by collecting statistics. When an agent A sends a query to another agent B and the query has not completed in some period of time, agent A faces the dilemma of whether to continue waiting or reissue the query (to agent B or to a different one). When some information is available about the probability distribution of the query completion time of the agents, it is possible to devise schemes for agents to reissue queries in such a way that the expected query completion time is reduced significantly. We design such schemes for various models of query-answering agents. Where meaningful, we take into account the cost of sending a query to an agent.