[OpenAFS] RE: read-write replication in other fs's

[Frank J. Cameron]

Microsoft's Dfs does not implement (at least what I consider to be)
true, full read-write replication.

http://www.microsoft.com/technet/treeview/default.asp?url=/technet/prodtechnol/windows2000serv/reskit/distsys/part3/dsgch18.asp

<quote>
FRS uses a "last writer wins" algorithm, which means that the last
update to a file or folder in a replica set becomes authoritative for
replication, regardless of the document version number or file size. It
does not merge changes; rather, the most recent version of a particular
file overwrites all older versions. This makes FRS well suited to
replicate files that are updated infrequently, such as product
specifications, software distribution points, and Web content. 

Files that contain information that is updated more frequently must
accommodate two scenarios: concurrent users and replication latency. 

      * User A and User B open the same 100-page document on different
        replicas. User A adds 100 pages and saves the document first;
        user B deletes 80 pages and then saves the document. The 20-page
        document that was saved last becomes the authoritative file. 
        
        FRS cannot deny file sharing or enforce file locking between two
        users who are writing to the same file on two different
        replicas.
/quote>

I can't find it now; but, I had another Microsoft document that
recommended not replicating data that changes frequently (especially
when changed by different clients).

However that the first failure case is what happens when AFS ignores
byte-range locks.

----------------------------------------
From: Nathan Neulinger <nneul@umr.edu>
To: openafs-info@openafs.org
Subject: [OpenAFS] read-write replication in other fs's

Someone recently suggested (at a training session) that microsoft's DFS
has fully redundant RW replication. Can anyone here corroborate or
refute that? More importantly - if it does, is the situation something
like:

        "Sure, it has RW replication in the same way that exchange has
active-active clustering. You're insane to use it cause it causes more
problems than it helps."