Are you guys concerned with decay at all? How do you know the file stored 2 years ago is still the same file? what happens if its become corrupted?
I recently read about facebooks cold storage and it got me thinking about it (https://code.facebook.com/posts/1433093613662262/-under-the-...) though for them they just want one copy total.
There are some potential attacks of putting false information into the git-annex repositories, that we use for tracking which clients are storing which files. We'll eventually need post-receive hooks to validate that pushes only change information about the client making the push. For now, if someone attempts this attack, we can revert their malicious changes after the fact.
Maybe for the time being, with published attacks.
But MD5's strength against attacks has been sufficiently dubious since early 1996 – before the founding of the Internet Archive – that experts have recommended against using it in new applications.
By 2009, the CMU Software Engineering Institute (responsible for CERT security notices) wrote:
Software developers, Certification Authorities, website owners, and users should avoid using the MD5 algorithm in any capacity. As previous research has demonstrated, it should be considered cryptographically broken and unsuitable for further use.
So they're worse than 'not ideal'. They're dangerously obsolete for the purpose of content-authenticity, and any reliance upon them (under hand-wavy conditions) serves to block the migration to available hashes that would actually work.
If you're making the backup to be sent through a wormhole to 1999, use MD5. If you're creating the backup for trust and availability through to end-of-year 2015, or 2025, or 2115, ditch MD5 ASAP.