Ask Slashdot: Linux Mountable Storage Pool For All the Cloud Systems?

An anonymous reader writes "Many cloud systems are available on the market like: dropbox, google, sugar sync, or your local internet provider, that offer some free gigabytes of storage. Is there anything out there which can combine the storage into one usable folder (preferably linux mountable) and encrypt the data stored in the cloud? The basic idea would be to create one file per cloud used as a block device. Then combine all of them using a software raid (redundancy etc) with cryptFS on top. Have you heard of anything which can do that or what can be used to build upon?"

Comments

Don't trust the cloud

[Anonymous Coward]

My residential internet connection via Comcast is fast enough today that I can pull files off of my server at home, "cloud" style.

I have two 2TB drives in RAID1, encrypted with whatever magic `cryptsetup' performs, with port 22 of my firewall forwarded to the server. SSH only accepts logins from me. I consider my data to be more secure and easier to access (it's literally seconds away from availability on any real operating system anywhere with internet access. Windows need not apply) than anything I could get from ZOMG TEH CLOUD. Only disadvantage is speed. I'm not gonna be shunting gigabyte plus files around like this.

Added bonus: easy to add users, easy to throw up a web interface, can do whatever you want with it, since you own the hardware (!!)

Pfft, cloud. I remember when it was called 'the internet'.

Now get the fuck off my lawn.

Re:There are several options here

[Omnifarious]

Tahoe sort of achieves this in an odd way. Directories contain hashes of the file they reference instead of an inode number. This means that a Tahoe node often doesn't even know who a file really belongs to, even though it knows its length.

The main issue with block storage is this...

Suppose you modify a data section of a file in a btrfs filesystem mounted on some kind of weird encrypted block device. There will be a whole tree of blocks that get modified, all the way up to the root node. All of these blocks have to be written before the root block is, and for a small file there will be several more blocks that need updating than there are data blocks on the file.

These two issues create a big synchronization problem and a lot of extra traffic.

In contrast, a good distributed filesystem protocol that's aware of individual files can send a single message that contains some kind of identifier for the file, and the new data it should contain. This message will often be smaller than a single filesystem block, and it will also usually be compressed before it gets on the wire. Much more efficient and while there are synchronization issues between updates to individual files, within a file there aren't any.