brandynwhite

This shows that they have to know who has access to each file, even if they don't want to. That is a privacy issue that other services are immune to because they don't actually make the claims that bitcasa does.

That is true but it changes the user experience and cost model pretty dramatically. This is designed to save all files including OS, so the de-duplication that they have would be only useful for storage and not bandwidth.

Flaw #3 and the 'shortsighted solutions' for #1 address this. Keeping quotas and logs leaks information about what you are storing. For example, if you keep statistics on the largest file a user has, then it dramatically reduces the number of files that the user's "true" file can hide in. This is a class of "side channel attack" (look that up for more on the subtle issues introduced).

So uploading was covered in #2 in the post. #3 is the downloading/user possession aspect. I am familiar with the DMCA which would reduce the scope the companies liability for #2; however, it does compel them to be compliant which hurts users (covered in #3) and there a variety of subtle situations where the DMCA wouldn't help them. Note that google makes copies of your music (in GMusic) instead of hash matching for similar reasons where Apple secured an expensive license to do this. Certainly Bitcasa doesn't have any special deals like that. So yes, to your point the DMCA reduces liability but I maintain that it is still a potential flaw (which is what the title of that paragraph is).

1. The issue isn't that bad guys exist and everyone has to deal with it. The problem is that for them to deal with it they have to sacrifice their user's privacy (quotas and stats open clients up to attack). If dropbox does that they don't subject their user's to any (more) security risks as they don't make such a strong assertion.

2. I am not a lawyer, but part of this post is to focus on the viability of the model. What I hope to come out of this is discussion on ways to reduce the legal exposure a company has in this privacy-centric model.

3. That is true but that isn't how copyright lawsuits generally work. For example, if you own a CD and they get you downloading something from bittorrent you'd have to argue that point in court. Even if you did nothing wrong that is more exposure than "client side encryption" would lead most people to believe.

I agree it is strong, though I wouldn't call it an attack piece. I will gladly update this as more information surfaces but I believe my assumptions are plainly explained and are likely to be true. The reason for my focus on them is they did many things exactly how I would have wanted them to, but it is dangerous to advertise a certain level of encryption that is generally understood and then provide something that has a variety of subtle flaws. I'll add an update to clarify #1 as that is important.

Edit: Convergent encryption link is fixed in the post (thanks for the tip)

Thanks! Let me know if you have any questions/comments. More documentation is here http://bwhite.github.com/hadoopy/

Yeah it's pretty loaded, hopefully there is something in there for everyone : )