Date: Mon, 23 Jan 1995 11:12:26 EST
From: "R. Alexander" <[email protected]>
> While we might be able to live with documenting to users to always
> do restores and achieves from one client for the whole campus, we
> can't live with trying to backup half the western
> (and perhaps eastern) world through
> user symlinks to other AFS cells. I am sure there is a programmatic
> way to skip these links...
>
> Both are problems we need to have addressed...otherwise ADSM
> will fail miserably here and IBM will be the target for a lot of
> carping.
> Is there anyone in ADSM development that can address these two problems?
I addressed the first problem in my previous response - a solution
is "cross-machine" principal support based on Kerberos in AFS, which is
supported in a prototype.
The second problem can be avoided by a check in ADSM client for AFS mount
points, and stop at those mount points when traversing the AFS tree. Crossing
mount points also has a danger, even if the mount points are not for foreign
volumes, because AFS mount points can form cycles. The prototype I mentioned
has the check included. The trade off is that one needs to get a list of the
mount points to be backed up as "virtualmountpoint". I have written a simple
script to produce such a list.
Shyh-Wei Luan
|