Heres the message:


Michael Poole <poole+@andrew.cmu.edu> writes:
>       I'm seriously considering writing a distributed file system for
> Linux, and wondering what I need to do (functions to call, functions or
> structures to expose, etc.) in order to make it work with the VFS layer.
> Is there any good example or documentation of this?
> 
> [Before people start saying that we don't need another distributed file
> system, I have my reasons: I have experience with NFS, AFS, and Coda.
> NFS has serious security problems on an insecure network, AFS isn't freely
> available, and Coda doesn't work right now, and might not work very well
> in a reasonable time frame.  If anyone knows of other distributed file
> systems available for Linux, or has constructive criticism or comments,
> I'd be glad to hear it.]

What exactly isn't offered by the existing file-systems that you need?

I'm guessing that it's the lack of security (from your comment above) 
- if so are you concerned about user identification/authentication or
about the fact that NFS transfers data in the clear.

The former could be solved by using DES or Kerberos authentication for
RPC, there is a secure rpc implementation for Linux (there's a copy at
ftp://sunsite.unc.edu/pub/Linux/system/admin/secure_rpc.tar.gz although
I doubt that it works in the newer kernels)

If you need session encryption what about NFS over TCP over ssh (youch!)

OTOH if you really want to implement a file-system what about looking at
a DFS implementation (I don't think that the Linux DCE offering includes
it).

I'm sure that you've put some thought into this and you haven't detailed
your requirements so it's difficult to second guess what's best for you.
However my advice would be

   1) Try to do what you want with existing components. This may require
      some effort to improve the quality of Linux ports. Advantage: you
      reduce the effort that you must put into the project as well as
      the time-scale.

   2) Look at whether you could meet your requirements by filling in a
      gap in functionality (e.g secure RPC for Linux). Advantage: by
      providing a generic component other people can make use of it for
      more than just file sharing.

   3) Consider whether doing a Linux implementation of an existing standard
      would meet your needs (e.g DFS for the Linux DCE). Advantage: there 
will
      be a greater number of systems that you can inter-operate with.

   4) Iff all else fails do something new. Advantage: you get to call the 
shots

Regards

Paul


-------------------------------------------------------------------------

Peter,

The thing that interested me most is the this person is already at cmu.  
Does he know that much about the project?  His comment on time frame 
seems pretty ironic, given he wants to start from scratch.  Since he 
seems interested in this sort of undertaking, maybe it would be good to 
know what objections he has to coda other than time frame (i.e. why not 
help with the existing project).

		Thanks,
		Dave


Dave 
http://insti.physics.sunysb.edu/~dstengle/
mailto:dstengle_at_insti.physics.sunysb.edu
finger for PGP key
Key fingerprint = E0 B0 5B EA 42 51 3B E2  E5 68 47 74 D4 FE CE 9D