Scheduling from the agents side.

In DIET, the scheduling works as follows (see Figure 7.1 for a representation of each step):

• A request is submitted to the Master Agent (step 1).
• The Master Agent forwards the request to the Local Agents and SeD s that it manages (step 2).
• The SeD s which dispose of the asked service return a CORBA response structure which contains an estimation metric vector (step 3).
• According to a default policy or a user-defined one, the responses from the SeD s are aggregated. Then the responses sequence is sent to the parent agent which aggregates all the results of its children (step 4).
• When the aggregated responses reach the Master Agent, it returns the aggregated list of all responses to the client (step 5).
• Finally, the client chooses the better server, according to the chosen aggregation method (step 6).

Figure 7.1: Scheduling steps in DIET.