Wow, this looks great, I'm going to install it at work tomorrow
to check it out. Seems like it could save us a lot of trouble as we
have a mix of OSX, Linux, Windows, Irix and Solaris boxes.
Maybe this is a stupid question but how about applications, I
assume they will need to be rewritten to handle AFS paths?
No, they're ordinary POSIX paths. The nice thing about AFS is
that there is a global namespace. My Spotify home directory is at
/afs/spotify.com/home/ehn; my KTH/NADA/CSC home directory is at
/afs/nada.kth.se/home/d00/d00-aeh; and my CERN home directory used
to be at /afs/cern.ch/user/e/ehn on any machine running AFS.
Authentication is handled by Kerberos. Very neat.
The ACLs are nice to.
That said, setting up the server is quite a project.
Thank you very much for clearing that up! Perhaps we could get
away with it on Windows too as we're using Cygwin (which is
partially POSIX compliant). I guess it depends a lot on how the
applications are written, the whole mounting scheme and mapping of
drive letters drives me crazy.
No need for Cygwin. There is a good OpenAFS client for Windows.
It seems to be the one that's most actively being developed, even.
I think they've received a Microsoft sponsorship.
But yes, in the native Windows environment, you don't get the
standard POSIX paths. You get \afs\domain.tld\directory...
Sucks.
@lemonad: I didn't mean
that AFS was bad and I don't really know if there are any good
options. It's just that it's heavily used at KTH and not really
used at all at LiU.
A couple of more question if you don't mind. I read it as AFS
handles big and little endian conversion. Do I also understand it
correctly that the conversion is done to fit the native platform?
That is, applications that aren't designed with that problem in
mind will work fine whereas cross-platform applications that
depends on querying the OS for its byte-ordering might not?
Sorry if that sentence makes absolutely no sense to anyone else
than me :)
I think what it means is that the bytes in a file will appear in
the order that you would expect on the platform where the file
system is mounted. In practice, it's never been an issue for me,
having used AFS on SPARC, x86, PowerPC and MIPS.
13 comments so far
Does it work on Leopard?
6 months, 1 week ago by sarnesjo.
Yes, very well. No issues so far. I'm running 1.4.7, which is the latest version in the stable branch.
6 months, 1 week ago by aehn.
Wow, this looks great, I'm going to install it at work tomorrow to check it out. Seems like it could save us a lot of trouble as we have a mix of OSX, Linux, Windows, Irix and Solaris boxes.
Maybe this is a stupid question but how about applications, I assume they will need to be rewritten to handle AFS paths?
6 months, 1 week ago by lemonad.
No, they're ordinary POSIX paths. The nice thing about AFS is that there is a global namespace. My Spotify home directory is at /afs/spotify.com/home/ehn; my KTH/NADA/CSC home directory is at /afs/nada.kth.se/home/d00/d00-aeh; and my CERN home directory used to be at /afs/cern.ch/user/e/ehn on any machine running AFS. Authentication is handled by Kerberos. Very neat.
The ACLs are nice to.
That said, setting up the server is quite a project.
6 months, 1 week ago by aehn.
Thank you very much for clearing that up! Perhaps we could get away with it on Windows too as we're using Cygwin (which is partially POSIX compliant). I guess it depends a lot on how the applications are written, the whole mounting scheme and mapping of drive letters drives me crazy.
6 months, 1 week ago by lemonad.
No need for Cygwin. There is a good OpenAFS client for Windows. It seems to be the one that's most actively being developed, even. I think they've received a Microsoft sponsorship.
But yes, in the native Windows environment, you don't get the standard POSIX paths. You get \afs\domain.tld\directory... Sucks.
6 months, 1 week ago by aehn.
@aehn So typical former KTH students to use AFS. :-)
6 months, 1 week ago by osks.
@osks I deferred writing just that yesterday. :)
6 months, 1 week ago by moonhouse.
@osks, @moonhouse: Sooo... I take it there are other equivalent or semi-equivalent options? ;)
6 months, 1 week ago by lemonad.
@lemonad: I didn't mean that AFS was bad and I don't really know if there are any good options. It's just that it's heavily used at KTH and not really used at all at LiU.
6 months, 1 week ago by osks.
Coda and Intermezzo are interesting alternatives (also from CMU), but they're far from being production-ready AFAIK.
6 months, 1 week ago by aehn.
@aehn, @osks: Thanks so much explaining further!
A couple of more question if you don't mind. I read it as AFS handles big and little endian conversion. Do I also understand it correctly that the conversion is done to fit the native platform? That is, applications that aren't designed with that problem in mind will work fine whereas cross-platform applications that depends on querying the OS for its byte-ordering might not?
Sorry if that sentence makes absolutely no sense to anyone else than me :)
6 months, 1 week ago by lemonad.
I think what it means is that the bytes in a file will appear in the order that you would expect on the platform where the file system is mounted. In practice, it's never been an issue for me, having used AFS on SPARC, x86, PowerPC and MIPS.
6 months, 1 week ago by aehn.