Olivier Mascia wrote:

><sidenote>As a side note, let's agree that cloning does not
>necessarily mean build a new clone DB on disk. The cloning process
>could pipe the pages to any other sub-process / driver / destination.
>Cloning to tape is not impossible. It just need a specific 'restore'
>tool to rebuild the db from the sequential stream of pages. That gives
>the usefull paradigm of backup image representing the state at
>backup-end time, while allowing to actually backup to some sequential
>device. I have another DB engine here working that way on 50 GB and
>larger DBs.</sidenote>
>
>Though as best I can remember some discussions I read carefully (at
>that time) on the devel list, NBackup would support incremental
>backup. Or was it some differential backup ? I remember of some backup
>version tag on pages. I might be wrong. Hopefully Nicholay will be
>back on line soon to comment by publishing full documentation.
>
>
>

I agree that a variant of clone could produce a stream that could be
reassembled into a database. It would only take a few lines of code,
impose no performance penalty, and the reassemble program would probably
fit on a single screen. It's the sort of feature that makes more sense
to implement than argue about.

All that said, what is the advantage of a serial medium? I gave up
watching tape prices long ago, but my impression is that cost per bit on
a cheap IDE drive rivals or exceeds the cost per bit of the most cost
effective tapes. A $129 250 GB disk (yesterday's Sunday flyer price)
plus a $15 tray is a might attractive combination. Even having conceded
that the implementation is nothing, does support for serial devices make
sense any more?

--

Jim Starkey
Netfrastructure, Inc.
978 526-1376