Abstract
In the last few years agents and multiagent systems have been strongly investigated by researcher communities all over the world because of the strong contribute and the great potential on software engineering and artificial intelligence fields. Since then, platforms and agent communication languages (ACLs) have been deployed in order to create environments where agents could be developed and implemented and in order to make agents interact with each other. Agent standardization has been carried out mostly by FIPA (Foundation for Intelligent Physical Agents) that introduced agent standards, from agent abstract architecture to transport protocols, from communicative acts to agent communication languages.
This work will mainly focus on agent communication, mostly on reliable group communication. FIPA ACLs consider only one-to-one interactions, they have not been designed for group communication purposes. However, we strongly believe that group interactions can be a milestone in agent societies as it is in human societies. For instance, teachers can talk to a wide audience of students (as a group) in a classroom; in the same way agents should be able to consider their message recipients not only as single entities, but as a group of entities as well.
Starting from previous works on group communication, we argue that group communication should be a complement to standard one-to-one interaction, rather than the only solution. We propose different forms of group interactions, considering agents as entities that (as humans do) have the freedom to choose the communication that better fits their needs. Different scenarios require different ways of transferring information, that is, in human societies face-to-face meetings are not the same as phone calls. In agent societies a similar argument should be supported.
Moving from our belief that agents should be able to choose their way to communicate, we argue that more on group communication should be done than standard asynchronous communication. Some scenario would certainly need stronger guarantees than the ones that can be achieved through an asynchronous infrastructure. We propose a virtually synchronous architecture, where group members receive joining and leaving members on their message queues as well as the same set of group messages between views. We introduce some new group semantics, we describe how an ACL message should be modified in order to support group interactions and we provide an architecture where agents can interact in a group oriented fashion.
Jade is a broadly used, FIPA compliant agent platform; we choose it as a base for our group oriented architecture. Even though it does not include any group capabilities, it provides an excellent plug-in service, which on one hand is really poorly documented (at least at the time this thesis work has been carried out), but on the other hand (once its features have been discovered) it makes quite easy for developers to design and implement a new service.