(Greg wrote:) > So a way forward that is not so objectionable architecturally and > addresses your usability concerns is to use some sort of python vfs > interface with multiple backends. Then at least it's not tahoe > reimplementing 12 wheels, but a larger python vfs community jointly > reinventing them only once (and perhaps a few that are done there > first). To me, this is an important distinction, where tahoe would > only have to support the generic module. This is much like p5-DBD and > 10 flavors of p5-DBB-Foo, except hopefully without the pain of > backend-dependent SQL leaking through.
Absolutely makes sense to have that work be generic and reusable, and not LAFS-specific. The related concern I have about LAFS for mass adoption is speed - mostly when users have a ton of files, so it'd just be important to make sure the latency for working with the backends doesn't go up much (which I don't imagine would be a problem). > Are you thining of this as users having access to a hosted service > (VPS?) running a node, where they lack administrative control, but want > to set up a backend using a cloud storage service for the ciphertext? > Or is this about users on locked-down corporate laptops? Or people who > just aren't comfortable? I'm thinking that people would want to be able to use storage they have already as a backend (or their company has or contracts for) without making a new commercial relationship, and even doing something with a docker instance or droplet on digitalocean where the users have to do 'Linux sysadmin' on backends is going to be a nonstarter for a lot of doit your-selfers. But LAFS's architecture already supports using untrusted backends, so removing the complexity of setting up LAFS nodes on the backend seems to make sene. And again, they can then do other things that they may care less or not at all about privacy for with those backend nodes. I understand that making it that easy may undermine some potential business for people offering LAFS-based services, but I think growing the whole pie and the ecosystem will benefit everyone. For example... Things like data center connectivity-driven delegated lease renewal and rebuild, and controlled proxy downloads of LAFS- uploaded content could still be value-adds offered by LAFS providers. I'm not suggesting the work done has been bad, or that anything is broken, but it was my first thought about LAFS as an even bigger newbie than I am now ("why can't I just use existing backends if encryption, rebuild, and lease management is done on the client side"). Avi _______________________________________________ tahoe-dev mailing list tahoe-dev@tahoe-lafs.org https://tahoe-lafs.org/cgi-bin/mailman/listinfo/tahoe-dev