On 2/7/2010 5:48 PM, Simon Wilkinson wrote:
>
> On 7 Feb 2010, at 20:14, Jeffrey Altman wrote:
>
>> I would prefer that the cache manager contact the file server in this
>> situation.
>
> Which file server? One chosen at random? And if so, in a world with 
> departmental fileservers, why should the client trust its information for the 
> whole of the cell?
>
> S.
>   

If we take the departmental file server model to the extreme, since the
'owner' and 'group' IDs associated with a file along with the access
rights on the file objects as viewed by the cache manager are obtained
from the file server hosting the volume, shouldn't the interpretation of
the client ID be obtained from the file server hosting volume as well? 
Such a model would provide the greatest flexibility for departmental
file servers to be administered independently of the rest of the cell. 
(Not that I am arguing for this extreme because such a model would
prevent volumes from being replicated across departmental file servers
and non-departmental file servers.)

On the other hand, if we take the approach that the PR database must be
uniform across the cell, why do we need to avoid sending requests from
the cache manager to the PR service?   The PR service is not like the
Vol service which is never contacted by a client.  While the PR service
is not contacted by the cache manager, it is contacted by other client
processes: *klog, fs examine, Windows Explorer Shell extension, etc."  
Is there a protocol reason to avoid sending requests to the PR service?

Jeffrey Altman




Attachment: smime.p7s
Description: S/MIME Cryptographic Signature

_______________________________________________
AFS3-standardization mailing list
[email protected]
http://michigan-openafs-lists.central.org/mailman/listinfo/afs3-standardization

Reply via email to