JXTA - Rendezvous Network

Rendezvous Network

The Rendezvous peers have an optimized routing mechanism which allows an efficient propagation of messages pushed by edge peers connected to them. This is achieved through the use of a loosely consistent network.

Each Rendezvous peer maintains a Rendezvous Peer View (RPV), a list of known rendezvous peers ordered by the Peer ID. There is not any mechanism to enforce the consistency of all RPVs across the JXTA network, so a given RPV can have a temporary or permanent inconsistent view of the other rendezvous peers. As soon as there is a low churn rate, that is, a stable network where peers don't join or leave too frequently, the RPV list of each peer will converge as each rendezvous peer exchange a random subset of its RPV with other rendezvous peers from time to time.

When an edge peer publishes an Advertisement, the index of this advertisement is pushed to the rendezvous through a system called Shared Resource Distributed Index (SRDI). After that, the rendezvous applies a Distributed Hash Table (DHT) function so that it can forward the index to another peer in the RPV list. For replication purposes, it will send this index to the neighbours of the chosen rendezvous peer in the RPV list.

The lookup process requires the use of the same DHT function to discover the rendezvous peer which is in charge of storing that index. Once the rendezvous peer is reached it will forward the query to the edge peer which published the advertisement and this peer will get in touch with the peer which issues the query.

If the DHT function cannot find a peer which is in charge of the advertisement then the query will be forwarded up and down the RPV list until a match is found, the query is aborted, or it reaches the limits of the RPV list. This process is called random walk.