On Tue, 21 Apr 1998, Per-Ola Mard wrote:
> Fellow AFS admins;
>
> I just wanted to thank you all for submitting your comments on the
> MR-AFS thread. A couple of thoughts where shared, of which I find the
> one from Andy Glew being very innovative! Creating the HSM functionality
> on a "disk-block" level is a cool idea! By making HSM transparent to the
> operating system connecting to that disk-device, you might be able to
> avoid all the fuzz on the filesystem/OS/application/network level. Nice
> thought! Made my head spinn an extra round.
>
> >From the comments of the people that responded to my initial query, I
> gather there _is_ a interest of getting the MR-functionality into the
> product. Officially. From Transarc. The AFS-vendor. We are still waiting
> for an "info-afs" comment on this from _Transarc_ though... Maybe next
> Decorum?
>
> We also got invaluable experience on the deployment of MR-AFS; create a
> new cell, do not extend your existing production cell with the
> MR-functionality.
This is really not necessary: we and also most other MR-AFS-cells
started with their standard AFS cell by just adding the new database
server "rsserver" and a MR-AFS fileserver.
You than may move volumes from your standard AFS-servers to the MR-AFS
server. These volumes, however, must be salvaged by the MR-AFS salvager
before the filesserver can attach them because the files need to get
additional vnodes where the access history of the file will be stored.
Who is anxious may pipe a "vos dump" into a "vos restore" instead of
doing a "vos move" in order to keep the original volume.
Another way is to convert a whole fileserver from standard AFS to MR-AFS
by installing the MR-AFS binaries and recreating the "fs" instance in
MR-AFS style. Then, of course, you also have to run the salvager before
the fileserver can start service.
There is also a way back: if you make sure that all files in the volume
have only the "local_disk" residency (that is the vicep-partition where
also the volume header is stored), you can do a "vos move" to a standard
AFS fileserver. The MR-AFS volserver recognizes that his partner on the
other side is standard AFS and doesn't send all the additional vnodes and
other fields. We did also "vos release" from MR-AFS to standard AFS
clones when we still had standard AFS servers.
- Hartmut
-----------------------------------------------------------------
Hartmut Reuter e-mail [EMAIL PROTECTED]
phone +49-89-3299-1328
RZG (Rechenzentrum Garching) fax +49-89-3299-1301
Computing Center of the Max-Planck-Gesellschaft (MPG) and the
Institut fuer Plasmaphysik (IPP)
-----------------------------------------------------------------
