Solving a problem

Assuming that the architecture described in Section 1.1 includes several servers able to solve the same problem, the algorithm presented below lets an MA select a server for the computation among those available. This decision is made in four steps.

• The MA propagates the client request through its subtrees down to the capable servers; actually, the agents only forward the request on those subtrees offering the service.
• Each server that can satisfy the request can send his performance and hardware information or an estimation of the computation time necessary to process the request to its ``parent'' (an LA) (via performance prediction tools: see Chapter 9).
• Each LA that receives one or more positive responses from its children sorts the servers and forwards the best responses to the MA through the hierarchy.
• Once the MA has collected all the responses from its direct children, it chooses a pool of ``better'' servers and sends their references to the client.