The Multi-MA

The Multi-MA is composed of all MAs running at the same time. The MA is able to connect the clients to others running MAs. Thus, each client knows only one MA, that is its access to the Multi-MA. Each MA publishes an advertisement with a lifetime in order to avoid clients or other MA to connect to a stopped MA. When it receives a request coming from a client, the MA submits the problem description to DIET via the MA it has itself launched. If the submission returns a DIET failure, the MA searches other MAs. Then, it forwards the client's request to other MAs. SeD references thus collected are merged and sent to the client.

The general algorithm of the MA is as follows:

• launch a new JXTA Peer,
• build an input pipe to listen to clients' requests or agents forwarded requests,
• create an advertisement including its input pipe reference allowing clients to connect to it back and publish it with a hardcoded lifetime,
• process each client or agent message by a thread :
  • if the source of the message received is a client,
    • call the MA with the packed problem and get SeD reference(s),
    • if any, send it to the client, else search other MA(s), forward the query to the other MA(s) discovered and send a response containing all SeD references thus received to the client.
  • if the source is an agent,
    • call the MA on the problem received and get SeD references found in its own DIET tree,
    • propagate the request to the other MA(s) (in order to find the fastest path to reach all the MA(s) on the network.
    • send a response including SeD reference(s) to the MA from which it received the request, and forward the responses from other MA(s) it has reached first back to the MA that reached first this MA.