Wikipedia article about Peer-to-Peer computing says - "Peer-to-peer (P2P) computing or networking is a distributed application architecture that partitions tasks or work loads between peers. Peers are equally privileged, equipotent participants in the application. They are said to form a peer-to-peer network of nodes."
While Wikipedia article for Multi-agent systems says- "A multi-agent system (M.A.S.) is a computerized system composed of multiple interacting intelligent agents within an environment. Multi-agent systems can be used to solve problems that are difficult or impossible for an individual agent or a monolithic system to solve. Intelligence may include some methodic, functional, procedural or algorithmic search, find and processing approach."
Although having difference in definition, both architectures involve cooperation between distributed entities.
How does one define the suitability of having a P2P system based architecture or a MAS architecture for a research problem?
P2P is a communication oriented terminology, whereas, MAS is related with how a physical entity can be represented in virtual computing, which in addition to other features, may involve a P2P communication function.
MAS usually involves artificial intelligence. If you need just communication, decentralization and distributed problem solving P2P is enough
P2P and MAS are not just different in definition, they are aiming at different objectives: In P2P, a distributed system is organized (by peers). Thus, you can build a system where dynamically new peers appear and other pears are removed. The system should continue in reliable state. P2P is more an infrastructural question of how to organize a distributed system. In P2P you can discover peers, connect to peers etc. The protocols are specialized for the setup and management of the infrastructure. In multiagent systems, the key concept is communication and coordination of the agents. The logical level here is rather high. In MAS you use communication which is based on messages which are sent from agent to agent. The agents are assumed to reason on the content and the type of messages and react with respect to this. The messages consists of two parts: one part has to be understood by any agent (the envelope: from/to, message type, ...) and an internal part, which is represented in some formal language (Prolog, RDF/S, ...) which has to be understood by the agents which take part in the coordination/communication process. So MAS is a mean for organizing distributed tasks on a content level and P2P is for putting together resources in a network.
Normally P2P purpose of collaboration is to obtain an individual benefit or goal, while MAS requires a higher degree of collaboration and coordination to achieve a global goal.
For example, in P2P file sharing program, I am mostly concerned about which files I can get and I will collaborate if that gives me benefits, like better download times or files; normally I even don't know what files are the other peers needing.
MAS needs more intelligence because it cares about a common goal and therefore requires a better coordination. It is not just about the communication but also how and what is communicated to be more efficient as a group. The communication in this architecture can be defined in different ways, either P2P or centralized.
I agree witj Ingo Timm. P2P could be used as an environment infrastructure for MAS such as a lower level of communication and support life. The agents could be developed as peer entities in this environment and, with a reasonable use of clustering development, its possible to deliver some low levels agent tasks for P2P platform and focusing agents on more intelligent issues.