Now, when one peer makes a request, it is possible that multiple peers have a copy of that requested object. Now the problem is how to get the IP addresses of all those peers. This is decided by the underlying architecture supported by the P2P systems. By means of one of these methods, the client peer can get to know about all the peers which have the requested object/file and the file transfer takes place directly between these two peers.

Three such Architectures exist:

1. Centralized Directory
2. Query Flooding
3. Exploiting Heterogeneity

Centralized Directory

• It is somewhat similar to client-server architecture in the sense that it maintains a huge central server to provide directory service.
• All the peers inform this central server of their IP address and the files they are making available for sharing.
• The server queries the peers at regular intervals to make sure if the peers are still connected or not.
• So basically this server maintains a huge database regarding which file is present at which IP addresses.

Now whenever a requesting peer comes in, it sends its query to the server. Since the server has all the information of its peers, so it returns the IP addresses of all the peers having the requested file to the peer. Now the file transfer takes place between these two peers.

The major problem with such an architecture is that there is a single point of failure. If the server crashes, the whole P2P network crashes. Also, since all of the processing is to be done by a single server so a huge amount of the database has to be maintained and regularly updated.

Query Flooding

• Now when one peer requests for some file, this request is sent to all its neighboring nodes i.e. to all nodes which are connected to this node. If those nodes don’t have the required file, they pass on the query to their neighbors and so on. This is called query flooding.
• When the peer with the requested file is found (referred to as query hit), the query flooding stops and it sends back the file name and file size to the client, thus following the reverse path.
• If there are multiple query hits, the client selects from one of these peers.

Gnutella was the first decentralized peer-to-peer network.

This method also has some disadvantages like, the query has to be sent to all the neighboring peers unless a match is found. This increases traffic in the network.

Exploiting heterogeneity

• This P2P architecture makes use of both the above-discussed systems.
• It resembles a distributed system like Gnutella because there is no central server for query processing.
• But unlike Gnutella, it does not treat all its peers equally. The peers with higher bandwidth and network connectivity are at a higher priority and are called group leaders/supernodes. The rest of the peers are assigned to these supernodes.
• These supernodes are interconnected and the peers under these supernodes inform their respective leaders about their connectivity, IP address, and the files available for sharing

KaZaA technology is such an example that makes use of Napster and Gnutella.
Thus, the individual group leaders along with their child peers from a Napster-like structure. These group leaders then interconnect among themselves to resemble a Gnutella-like structure.