[OpenAFS] questions on AFS

Martin Schulz schulz@iwrmm.math.uni-karlsruhe.de
05 Dec 2001 09:28:59 +0100

"Srinivas, Mohan" <mohan.srinivas@digital.com> writes:

> Hi everybody,
> can any one explain these questions?

Well, I am not very deep into these issues, but I try to explain what
I know up to now.

> 1) Does AFS supports file level migration? If not what 
> is the reason?

What das file level migration mean? Changing the physical location a
file is located (its read-write location)? If so, the answer is no. 

AFS uses an abstraction called volumes, which can be regarded as
disjoint subsets of the files in the filesystem. These volumes are
treated as an entity for administration, such as quota, backup or -
you guess it - location (that is, the file server and the
corresponding partition).  

You can move these volumes between servers or partitions, but not
single files. Clearly, you could set up a volume per file...

> 3) How about use of database over AFS? is it advantageous or 
> disadvantageous to use AFS for this application?

If any client is writing to the database, the changes are not
guarantied to be immeadeately visible to all other clients. This is
guarantied upon the closure of the file. Furthermore, the other
clients have to refetch the whole file if they need the updated file

I don't think large distributed databases should be realized by file
sharing via AFS, I suggest using tools made for that task. I know of
Adabas D to support that, there are surely others as well.

If you merely distribute some .dbm files that are rather seldomly
updated, then AFS will surely be convenient.

> advance thanks and happy christmas to everybody,

Ooops, so late already? 

Martin Schulz                             schulz@iwrmm.math.uni-karlsruhe.de
Uni Karlsruhe, Institut f. wissenschaftliches Rechnen u. math. Modellbildung
Engesser Str. 6, 76128 Karlsruhe