Welcome
Visit us on #arf @ irc.freenode.net
About
We created this site in order to gather information and concept ideas for a new network file system / file transfer protocol. We however tend to look at our idea rather as an abstraction layer for remote file systems.
Eventually we decided for arf(Access Remote Files) acronym since everything else which was sensible was already in use. :) Oh, and we get a nice mascot:
Why?
FMD - FTP must die!
Also other alternative protocols like sftp do not deliver a satifactory performance. We intend to fix the issues with them. All of those systems also greatly lack on features …
Desired filesystem properties:1
- A clear well documented protocol for everyone to implement.
- Client-server model running over a single(unlike FTP, which uses many connections) SCTP connection.
- Security accomplished by layering the connection over the proofed SSL mechamism(unencrypted transfers should also be possible, but authentification should always be encrypted)
- Performant! Wasting bandwidth and not thinking about the worst cases is bad; we intend to assume the file system working even on google data centres. (not a bit too much? :D)
- Flexible! On the bottom the filesystem will _NOT_ be designed explicitly for Windows or Unix. It will allow general file transfer capabilities + other optional capabilities a system might offer. Nothing like that on the lowest layers though.
- Allowing extensions, a bit like reiser4, allowing for example a client to browse files as directories, allowing the server to unpack compressed files, allowing the client to browse archives without first downloading the whole file, or the server sending remote updates made to the tree.
- Responsive! Achieved by asynchronous design and non blocking operations. Also huge operations might be divided into chunks.
- Both mountable(via FUSE for example) and connectable the standard way.
- Transaction based - like deleting 1000 files requires only sending one query.
the above order is random
Site contents:
Todo:
- Remote filesystem model - Polish, make it readable, possibly rethink.
- Control draft - please provide your proposals.
- Atomicness - How will it work, how should it play with transactions?
- Transactions - How much sqlish, to what extent?
- Message flow
- Other TODOs
People currently involved:
Want to get involved ?
page revision: 31, last edited: 16 Sep 2007 05:30
