Efficient Search Mechanisms for Learning Mobile Agent Systems

In order for mobile agents to achieve their goals they must possess information about the location of objects in their environment. Given that mobile agents are often intended to cope with highly dynamic situations, it is clear that some mechanism should be employed to make sure this information is correct. This paper investigates the difficulties in providing location database support for mobile agent systems, particularly as regards the expenses (such as time and bandwidth) incurred during search for objects or the storage of their location. A technique is presented that, for a specific set of cases, optimises network resource consumption through dynamic selection of agent interaction protocols. This process is made possible through learning on the part of static knowledge agents (acting as location databases), which can then predict the likely length of a given search procedure. It is these predictions combined with information about the network environment that allows quantitative comparison between different search protocols.