Joshua is an attempt to write a (globally) distributed, resilient, read-through file server. Don't use it right now, I'm just trying it out.

The archtypical use case, and the one I'm attempting to cover, is a web app with servers in multiple data centers that accepts uploads from users. If you upload a file to one data center, it should be available for reading or deleting in any other DC.

• Apparent durability. Once data has been written to any node, it MUST be available from any node.
• Automatic eventual consistency. If you upload a dozen files to one node and wait a little while, all nodes MUST eventually be consistent without any further human action.
• HTTP read access. Consumers MUST not need to touch the file contents once it's uploaded.
• Self-bootstrapping. If a new node comes online, it MUST notify the cluster and populate itself.
• No database dependency. For metadata, it SHOULD not be necessary to run a local database or data store outside the server process. (A database-backed datastore is allowed.)
• Protocol-based. It SHOULD be possible to implement the protocol in another language and join the cluster.

• Load balancing. Use anycast between DCs. Use a load balancer within a DC.
• Updating objects. Only create and delete operations will be supported.

• Web-based status and monitoring. View cluster status over the web.
• Pluggable storage backends. Starting with the file system, but MongoDB is an obvious place to go.
• Performance. Obviously.