If you use hashes of the content itself for your UUIDs, you'll (a) get deduplication and data consistency checking for free and (b) have basically implemented (a subset of) git that uses S3 backing instead of a local filesystem directory :)