Abstract for Zakaria Maamar and Nader Troudi Name: Zakaria Maamar Organization: University of Laval E-mail: maamar@ift.ulaval.ca Go to the bio Name: Nader Troudi Organization: University of Laval E-mail:

Migrate or Not Migrate, That is the Question Zakaria MAAMAR and Bernard MOULIN {maamar,moulin}@ift.ulaval.ca Computer Science Department and Research Center on Geomatics Laval University, Ste-Foy, QC, G1K 7P4, Canada

The mobile agent paradigm is a new alternative to traditional client/server computing. At the basis of this paradigm is the idea of decomposing the agent program into a set of portions of code that can be executed in a sequenced manner on different system hosts. Before a portion of code can be executed, the agent migrates to the appropriate host across the network. Generally, the decomposition of the code is based on the availability of resources that are needed for the agent for executing the various portions of code.

There exist several projects in the field of mobile agent (Agent Tcl, MOLE, TACOMA, Telescript,~etc.), all of which have focussed on the description of an agent operations when it decides to move from host to host. Hence, the agents are reactive and do not have the ability to reason about their environments and to infer new knowledge.

At Laval University, our research group under the supervision of Prof.~B.~Moulin, has started a new project that deals with mobile agents, enhanced with planning, coordination, and negotiation mechanisms. Our objective is to use cognitive agents that can evolve according to their own knowledge and decide when migration is preferable to requesting agents for distant processing.

At this moment, in order to assign planning capabilities to an agent, we use several concepts, such as plan, service, resource, ability, and place. We are also investigating about operations of invitation, which are the opposite of migration (an invitation is formulated by an agent that requires resources to another agent that has the required resources). An agent is located in a place that belongs to a host on the network. The agent has the ability to plan its tasks in order to achieve the services that are invoked by users. Furthermore, the agent decomposes its plan into a set of subplans, according to the task's complexity. If an agent cannot carry out a subplan on its own, it can request services from other agents located in other hosts (local or distant); the requested agents are viewed as resources. Depending on the issues of resource cost and migration risk, an agent can either decide to move to other hosts or to request services from other agents or even to invite certain agents to its own host. In this latter case, the invited agents have also to evaluate the possibility of migration in terms of earnings and risks.