Thought Torrent - Wuala

Wuala is another attempt at end-user distributed storage. It provides 1GB of storage to start with that's provided by the company's server. For additional storage, you can invite up to 15 friends and get 1GB of storage for each one that joins. Beyond that, you can trade storage on your computer(s): (# of GBs)*(percent of uptime). Also, your download speed is proportional is adjusted based on your (computers) reputation. Unlike Freenet, they've traded anonymity in favor of accountability. Files and directories are encrypted (both data and metadata) with private, limited sharing (both individuals, e.g. alice and bob, and groups), and public systems.

Notably, it's alpha and read-only at the moment ... [Edit: It's read-only in the sense that you can't edit files. You can create and delete files and directories]

Files are broken up into chunks. Instead of duplicating chunks of a file a lot (like 25 times), they use erasure codes (bringing it down to 5 or 6, assuming 25% uptime per-node).

If a lot of requests come in for the same chunk within a short period of time, the requests are redirected to other nodes that downloaded the chunk recently (providing bittorrent-like swarm distribution).

Access control is provided using a structure called cryptree. It uses lazy revocation, so once someone gains access to a file they can always read the last version they had access to; on the other hand, this means that multiple files can be encrypted with the same key and updating one file doesn't require that all the files be re-encrypted. Also note that they could just save the file itself anyway.

Cryptrees use asymmetric crypto (RSA with 2048-bit keys) to make the connections directed. Lazy revocation is implemented via key regression; when updating a file or folder encrypted with a key, K, a new key, K', is used and the pair (K',K) has the property that K' can be used to compute K, but K can't be used to compute K'.

Public files are indexed on the central server, private and shared files are indexed locally.

Routing is handled using the usual distributed hash tables with random links ... the only thing that might be interesting is that in responses to requests, nodes add some information about nodes they're aware of. This is intended to help discover new(ly online) nodes quickly.

Fairness is ensured by random auditing and reporting, to prevent nodes from lying about how much they're storing or how much time they've spent online. A large number of nodes should have to lie to break the system (paper); presumably reports are signed to prevent spoofing. However, I don't see any protection against sybil attacks. The paper simply states that they're handled by a different part of the application.