Like Gnutella, Freenet [10] is a peer to peer network of servers. When a user wishes to request a document, she hashes the name of that document (where she gets this name is outside the scope of Freenet) and then queries her own server about the location. If her server does not have it, it passes the query on to a nearby server which is ``more likely'' to have it. Freenet clusters documents with similar hashes nearby each other, and uses a routing protocol to route queries ``downhill'' until they arrive at the desired document.
Freenet bases document lifetime on the popularity of the document: frequently requested files get duplicated around the system, whereas infrequently requested files live in only a few places or die out completely. While this is a valid choice for a system that emphasizes availability and efficiency, it precludes certain uses of the system. For example, Yugoslav phone books are currently being collected ``to document residency for the close to one million people forced to evacuate Kosovo"[32]; these phone books might not survive a popularity contest.
Freenet explicitly sets out to provide anonymity. Their goals include both sender and reader anonymity, as well as plausible deniability for servers - the notion that a server does not know the contents of documents it is storing. They provide this last, which we called isolated-server document-anonymity, by referencing files by H(name) and having users encrypt the documents themselves with name before inserting them. This means that anybody who knows the original name string can decrypt the document, but the server storing the document is unable to invert H(name) to determine name.
Freenet has a similar potential flaw with publisher- and reader-anonymity to Gnutella, due to the presence of the TTL and Depth (comparable to Hops) fields in the Freenet message headers. Freenet takes steps to avoid the problems of Gnutella's Depth and TTL headers by randomly assigning values to both fields, so that a depth of 1 does not necessarily mean that a request originated with a given node. Packets with TTL 1 are randomly expired or forwarded onwards.
Document requests are also sent through the caching-enabled network (rather than peer-to-peer as they are in Gnutella). Because of these measures, Freenet has ``more'' anonymity than Gnutella provides.
Further, statistical attacks similar to those described in the Crowds [36] paper might work to pinpoint the location of a given reader or publisher; caching provides protection against this since the network topology for a given document changes after each request. These attacks need to be analyzed further.
Another attack to consider is that simply doing a request from a strict political environment will cause a copy of the data to migrate into that jurisdiction, giving law enforcement more reason to shut those servers down. On the other hand, it may well be that such environments will close down servers without requiring ``sufficient cause'', so the political and legal arguments are meaningless.