I don't understand, do you have access to a CDN or you're saying that soundcloud and ch9 have a CDN?
I think I saw a post somewhere saying Channel 9 uses or has moved to a CDN on Azure. I think it was when many of us were experiencing deja-vu seeing posts from a few months back
I don't think they're stored in a database at all. If I were to create such an architecture, I would use a RDBMS for storing metadata and just a file system to store the actual large files.
I guess for me it would have to be using a tried a tested metadata system. That is the part that I am most concerned about. It's not that I am scared of innovating, but this is something I don't want to have to have to write and test from scratch. Is there a system in wide use?
My understanding of NoSQL systems is that they're used mainly where you have to process massive (upwards of millions of items) amounts of loosely typed data, not where you have a small amount of large files...
I have only looked into this briefly, but I have more questions than answers at the moment.