Gnutella

Each gnutella node is a "servent" (agent?).
New nodes join by communicating with at least one existing node. A handful is better for connectivity.
When A wants to do a search, it sends the query to B. B returns any matching documents, records the request, and passes it to C. C will return the replies which B will then forward to A.
- Each request is given a unique ID, so duplicates are dropped.
- Each request also has a Time To Live (TTL), so it will eventually die.
- All data uses http, and port 80.
It has some problems:
- A search with TTL=10 will reach a million hosts. This does not scale well.
- A TTL horizon imposes a horizon on all users. I can only talk to my near neighbors, no matter how many times I try.
- Nodes can provide false results.
- There is not way to know which node the data came from.
Could these problems be solved by making the nodes into selfish agents?