Editing Hyphanet (section)

===Protocol===
[[File:Freenet Request Sequence ZP.svg|thumb|A typical request sequence. The request moves through the network from node to node, backing out of a dead-end (step&nbsp;3) and a loop (step&nbsp;7) before locating the desired file.]]

The Freenet protocol uses a [[key-based routing]] protocol, similar to [[distributed hash table]]s. The routing algorithm changed significantly in version 0.7. Prior to version 0.7, Freenet used a [[heuristic routing]] algorithm where each node had no fixed location, and routing was based on which node had served a key closest to the key being fetched (in version 0.3) or which is estimated to serve it faster (in version 0.5).  In either case, new connections were sometimes added to downstream nodes (i.e. the node that answered the request) when requests succeeded, and old nodes were discarded in least recently used order (or something close to it). Oskar Sandberg's research (during the development of version 0.7) shows that this "path folding" is critical, and that a very simple routing algorithm will suffice provided there is path folding.

The disadvantage of this is that it is very easy for an attacker to find Freenet nodes, and connect to them, because every node is continually attempting to find new connections. In version 0.7, Freenet supports both "opennet" (similar to the old algorithms, but simpler), and "darknet" (all node connections are set up manually, so only your friends know your node's IP address). Darknet is less convenient, but much more secure against a distant attacker.

This change required major changes in the routing algorithm. Every node has a location, which is a number between 0 and 1. When a key is requested, first the node checks the local data store. If it's not found, the key's hash is turned into another number in the same range, and the request is routed to the node whose location is closest to the key. This goes on until some number of hops is exceeded, there are no more nodes to search, or the data is found. If the data is found, it is cached on each node along the path. So there is no one source node for a key, and attempting to find where it is currently stored will result in it being cached more widely. Essentially the same process is used to insert a document into the network: the data is routed according to the key until it runs out of hops, and if no existing document is found with the same key, it is stored on each node. If older data is found, the older data is propagated and returned to the originator, and the insert "collides".

But this works only if the locations are clustered in the right way. Freenet assumes that the darknet (a subset of the global social network) is a small-world network, and nodes constantly attempt to swap locations (using the [[Metropolis–Hastings algorithm]]) in order to minimize their distance to their neighbors. If the network actually is a small-world network, Freenet should find data reasonably quickly; ideally on the order of <math>O\big([\log(n)]^2\big)</math> hops in [[big O notation]]. However, it does not guarantee that data will be found at all.<ref name=Clarke2010>{{cite book |last1=Clarke |first1=Ian |title=Private Communication Through a Network of Trusted Connections: The Dark Freenet |date=2010 |url=https://freenetproject.org/papers/freenet-0.7.5-paper.pdf |access-date=15 September 2015 |ref=Clarke2010 |archive-date=1 December 2017 |archive-url=https://web.archive.org/web/20171201033416/https://freenetproject.org/papers/freenet-0.7.5-paper.pdf |url-status=live}}</ref>

Eventually, either the document is found or the hop limit is exceeded. The terminal node sends a reply that makes its way back to the originator along the route specified by the intermediate nodes' records of pending requests. The intermediate nodes may choose to cache the document along the way. Besides saving bandwidth, this also makes documents harder to censor as there is no one "source node".