ADSM-L

Re: "Point-in-time-restore"?

1994-02-01 15:55:16
Subject: Re: "Point-in-time-restore"?
From: Paul Zarnowski <[email protected]>
Date: Tue, 1 Feb 1994 15:55:16 EST
On Tue, 1 Feb 1994 14:03:14 GMT Smith, Dean said:
>Why turn BACKUP/RESTORE into a function which is addressable through ARCHIVE?

I've been watching this and trying to resist getting pulled in, but I just
can't resist.

Dean, I guess I don't understand your unwillingness to concede that
there might possibly be a need to restore a disk (or directory or
filesystem) to it's state at a point in time, without knowing in
advance what that point in time might be.  A few possible reasons have
been listed to illustrate the point, and you have attempted to dismiss
all of them.  We could doubtless come up with dozens of additional
scenarios where this feature would be useful.  Perhaps you will remain
unconvinced until you come up against one of these situations yourself
and experience it firsthand.  The whole point of backing up is to
protect yourself against the unexpected.  If you could predict
everything, then you wouldn't really need a backup system.

In my mind, the ARCHIVE function should not be contorted into solving
this type of backup problem.  ARCHIVING should be used for storing
something that you want to keep around for awhile.  Using ARCHIVE for
taking numerous snapshots of filesystems, just in case you need them,
is wasteful of network bandwidth, database space, and storage capacity.
There is no need to make a copy of an entire filespace, just in case
you might need to restore it to a point-in-time, when only a few files
have changed.  Incremental backups are very efficient at only backing
up files which have changed.  The problem is, the ADSM backup
philosophy does not accomodate point-in-time restores very well.

The challenge, in my mind, is to figure out how point-in-time restores
could be accomodated (if at all) within the ADSM backup framework, without
severely contorting the existing framework.  I like the existing framework
for most things, but I do see a real need for point-in-time restores.
Rather than saying "the need does not exist", I think it would be more
useful for this group to talk about how the need might be accomodated
within the current ADSM framework.

Perhaps it cannot be accomodated by ADSM.  But perhaps it can.

..Paul
<Prev in Thread] Current Thread [Next in Thread>