On Fri, Sep 16, 2005 at 09:39:41PM +0200, Viktor Pracht wrote:
>
> If the SVN box dies, and there are no backups, then Traversal Technology
> has lost all its documents, can't produce the next generation of the
> chip, investors go away, Timothy and his partners have no money to start
> from scratch again, the dream of OpenSource-friendly hardware is over,
> Linux slowly becomes irrelevant because of bad hardware support,
> Microsoft expands until Bill Gates controls every government in the
> world...
> But at least the number log is safe because it was backed up
> separately ;)
>
> Traversal will have to find some way to back up all the specs and
> documents anyway. Having the number log in a different backup system
> does not add any redundancy (the reason why one normally has multiple
> backups), and instead increases the amount of code where a bug can
> seriously damage Traversal.
You're right that this problem will also require attention before
long.
Revision control for paper originals of engineering drawings is a
well-established discipline. I've worked at a couple of companies that were
struggling to develop a safe system of paperless drafting, and I didn't see
anything that impressed me as adequate over the long haul.
The situation for revision-controlled released drawings is somewhat
different from number logs, though. Fundamentally, CAD drawings are
discrete files, which are revised in a batch according to a signed ECO, and
then checked by hand for correct completion of the ECO, and signed off as
completed. My last company plotted out a new hardcopy original, which was
physically signed and filed in the Drafting drawers. That was our
protection against a computer disaster or system obsolescence. There are
more details, but that's the basic idea.
Drawings don't accumulate transactions on-the-fly, as a number log
or a conventional data base does. There is no need for 24/7 write access by
all project members. Only a few drafters and clerks who can be depended on
to follow ECO discipline have write access to the release file tree.
Unreleased drawings, on the other hand, are not under this kind of
tight discipline. They're in active development, and all sorts of people
have access to working drafts. A CVS or SVN tree is ideal for that kind of
work (although the places where I worked perversely used Source Safe
instead). The need in active development is to maintain a clear and
up-to-date picture of the work in progress, not to assure failure-proof
archiving for decades, future-proofing, and protection against unauthorized
revision of ECO-controlled originals. If something happens to a SVN archive
that's used mostly for active development, an error isn't catastrophic. The
project team is still around, their ideas are fresh in their heads, and they
can reproduce a few days of recent work without too much trouble. This is
why an engineering and drafting group makes a clear distinction between
released and unreleased drawings, and has very different revision and
archiving requirements for the two cases.
Now, possibly, some combination of software design principles and
physical archive protection might be useful in both types of problems,
drawing archiving and distribution, and number log maintenance. I'm not at
all sure. I think I'll lurk and learn while you talk over possible
techniques.
One tidbit is that I had a conversation once with a 3M applications
engineer, about the storage lifetimes of different machine-readable media.
He told me that if you buy good quality material, and store it in stable
temperature and humidity conditions, a floppy disk should last at least 25
years, and a CD-R should be good for 100 years.
This suggests to me that when we get around to writing an ECO
procedure and releasing product designs for production, we might want to
periodically batch recent ECOs and the resulting revised CAD files on a
regular schedule, and burn them to CD-Rs for permanent off-site storage.
Restoring the full set of CDs in order would recreate the release tree, up
to the last archive burn.
Under a scheme like that, the worst-case risk would be confined to
having to re-execute all ECOs since the last archive burn, working from the
hard-copy original ECOs. (An ECO cover sheet has to be a hard-copy document
anyway, because that's where the department representatives put their
approval signatures.) Actually, we'd probably retrieve the updated CAD files
from the designers' private drives or the SVN repository, and only repeat
the completion checking before copying them into the release tree.
_______________________________________________
Open-graphics mailing list
[email protected]
http://lists.duskglow.com/mailman/listinfo/open-graphics
List service provided by Duskglow Consulting, LLC (www.duskglow.com)
