[OpenAFS-devel] Win2K AFS server config help; setting up harmless new test Unix server

[Chris Huebsch]

Hi,

On Tue, 30 May 2006, Winnie Lacesso wrote:

>  On an AFS client, microsoft or Unix, directories
> seen partially do & partially don't match what volumes
> are seen on server.

How do you "see" volumes on the client?

On the client you see directories. Starting from /afs (which is normally
populated with the content of root.afs) and going through so called
"mount points", which point to the root-directory of other volumes.

>  So there is a mapping somewhere between server
> volumes & what clients see..

In a kind of speaking - yes. But do not mistake AFS for a filesystem
like NFS or Microsoft Shares (SMB). AFS-Server maintain their data in an
own data space located in /vicep[a-z] (and even /vicep[a-i][a-z]).

You cannot use ordinary filesystemtools like "cd" or "ls" in the
/vice-paritions to see what the clients see.

>  Where is that config file, that mapping of what
> volumes on the afs server get exported or viewed-as on
> client machines? I can't find anything like it on the
>  AFS server.

The mapping you are looking for, are mountpoints in the filesystem
inside the AFS-Volumes themselves. They are a kind of "symbolic link"
linking directory names to volume names. If a client traverses such a
mountpoint, the AFS-client automagically queries the AFS-Servers for the
content of that volume. (This indirection is one of the most powerful
concepts in AFS, because this allows volumes to be migrated between
servers without being noticed by the client in their filesystem-view.)

> - I'd like to setup a new test Unix AFS server,
> getting ready to migrate off win2k. Can someone
> itemize all the openafs rpms to install on Unix server
> & in what order? All the install docs are ancient &
> assume you have an IBM AFS install CD. Is there a
> modern install doc?

Speaking of RPMs I guess you use a Redhat-ish distribution?

You need openafs and openafs-server for a server-only installation.
If you also want to have client functionality (which means you can
access /afs) on your server, you need openafs-client RPM and the
openafs-kernel-module RPM too. The openafs-kernel RPM has to match your
running kernel version. You might to be forced to rebuild this module
yourself.

The order of the packages is resolved automatically by your rpm-tool.

>  - Is it safe to setup a new Unix test server with
> some dummy/test cell name, which server is either
> unknown to the planet, or known only as
> un-production/off-limits? I don't want to cause ANY
> problems/interference with  production AFS!

You can use whatever cellname you want with your Unix-Server. You do not
have to publish the name/ips of your AFS-Servers. If you do not do this,
no one will get problems. But if you use example.com for your test-cell,
there is no problem to be expected at all.

I suggest, that you at first try to install a test environment to get
used to the new concepts of AFS. After that you can setup a second
AFS-Server to your current installation and move all data online to your
new server without interrupting operation in your cell.

This documentation (although for gentoo-linux) is very good:
http://www.gentoo.org/doc/en/openafs.xml

Just read Chapter 5 ans skip the compilation-step.

Greetings


Chris
-- 
Chris Huebsch    www.huebsch-gemacht.de | TU Chemnitz, Informatik, RNVS
GPG-Encrypted mail welcome! ID:7F2B4DBA |   Str. d. Nationen 62, B204
Chemnitzer Linux-Tage 2007, 3.-4. Maerz |       D-09107 Chemnitz
     http://chemnitzer.linux-tage.de/    | +49 371 531-931118, Fax -1803