[OpenAFS] Migration from Transarc AFS to OpenAFS

[Wheeler, JF (Jonathan)]

Thank you to the list for the previous helpful advice about "Out of date
entries in the /afs directory".  Now for the next question !

I am planning the migration of our cell from Transarc AFS to OpenAFS and
would welcome comments on the following proposal.  At present I have 1 AIX
system with 2 RAID arrays attached (call it N for Now) which runs Transarc
AFS.  I intend to install 3 Linux systems each with 1 RAID array attached
(call them A, B and C).  My plan is that once I have installed OpenAFS on
the Linux systems, I configure the cell locally (on the servers) to consist
of N, A and B which will allow me to move all the volumes from N to A and/or
B (I think I could also add C as long as I configure that system to be a
fileserver only).  At the same time I will announce via the CellServDB
mailing list that our cell will consist of nodes A, B and C rather than node
N.  The idea is that other cells running with the old CellServDB will
contact the cell via N and cells running with the new CellServDB will try
and contact the cell via A, B or C (if they try to contact C they will time
out and try one of the other nodes - A or B).  Of course, later I will
reconfigure the cell locally to use A, B and C and switch off N.

I am doing things this way to try to ensure that the cell stays active
during the migration and I understand that 3 servers in a cell is a good
number to have to ensure a quorum is maintained.  I expect that there will
be times when one of the nodes will require a reboot; at that time I assume
that the quorum will be lost until the rebooted node is running AFS again,
but the only problem will be the inability to update any of the AFS
databases.

Does this seem to be a workable plan ?  If not, what are the flaws ?  I will
have a separate test cell which I will use to make sure that I can install
and run a cell using OpenAFS.  Many thanks in advance.

Jonathan Wheeler
e-Science Centre
Rutherford Appleton Laboratory