[OpenAFS] Current comparison of NFSv4 with AFS.

Jim Rees rees@freebsd.org
Mon, 23 Jul 2007 08:49:18 -0400


Jesse W. Asher wrote:

  Anyone have any real world experience they are willing to share on how 
  AFS compares with current implementations of NFSv4?

I use both every day.  I haven't done any kind of formal analysis but I'd be
happy to share my impressions.  These are gross generalizations and personal
opinion so don't bet the company on this.

  1)  WAN performance

Raw transport performance should be better for nfs because it uses tcp
transport.  But afs has an edge in some cases because of the on-disk cache.
On my trans-atlantic link afs seems to take a long time to get going, but
once it does it's snappier than nfs.

  2)  Global namespace (transparency of file/directory location)

Afs wins.  I pushed for global namespace in nfs but lost, because people
feel backward compatibility to v3 is more important.  There is a proposal
now for global name space in nfs using dns srv records but I don't think
it's implemented by anyone.

  3)  Caching

Afs is better if you prefer consistency, nfs is better if you prefer
performance.

  4)  Security
     a)  Encryption

Afs uses an ancient relative to des.  That's fine for what I do but a bank
or government office might prefer nfs, which can use several different types
of strong encryption.

     b)  ACLs

I find nfs acls to be confusing and hard to use but that is mostly because
I'm used to afs acls.  Nfs has "kitchen sink" acls that try to be a superset
of every acl scheme ever invented.  One big advantage to nfs is that you can
put an acl on an individual file.  In afs all files in a given directory
have the same acl.  This is a pain in ~ and ~/.ssh, for example.