Re: Dropping ACID - was Re: does a table always need a PK?

"Christopher Browne" <cbbrowne_at_acm.org> wrote in message news:bircns$ccakp$1_at_ID-125932.news.uni-berlin.de...
> After takin a swig o' Arrakan spice grog, Jonathan Leffler <jleffler_at_earthlink.net>
belched out...:
> > Christopher Browne wrote:
> >
> >> Alfredo Novoa <alfredo_at_ncs.es> wrote:
> >>>BTW, does anybody knows if Date and Darwen are considering to drop
> >>>transactions in the next revision?
> >> Drop them? Or "drop them into the book"; they only made limited
> >> comment about them in the last revision. (Which doesn't imply that
> >> transactions are irrelevant, just that they weren't relevant to the
> >> points they were trying to make about relational databases.)
> >
> > Drop them is the question - see the comments in Date's 8th Edition
> > (Section 16.10 - Dropping ACID).
> >
> > I don't know is my answer - but maybe. Of course, the replacement is
> > the 'multiple assignment' operation (including multiple insert or
> > delete or update operations), and it has a number of attractions. I
> > would certainly love to have that available. I'm not sure how much
> > you need transactions if you have 'multiple assignment'; nevertheless,
> > at the moment, I think I still like the idea of nested transactions. I
> > need to re-reread the material cited.
>
> Very interesting.
>
> I haven't a handy copy of _Date_, as the local libraries aren't big on
> high end texts.
>
> That is a _very_ curious proposal; it surely bears some comment.

Christopher, if you look back a few months in c.d.t, you will see some posts under the threads 'Transactions: good or bad?' and '"Transactions are bad, real bad" - discuss' where I was defending my personal position that transactions are not compatable with the logical level of the relational model.

I won't speak for Chris or Hugh, but I'd be surprised if transactions aren't at least demoted somehow in any further edition of TTM.

> It would be most ironic if relational databases got around to dropping
> ACID just when the MySQL people try to get their implementation
> mature. :-)
>
> The issue of how strongly concurrent transactions need to be like
> their serialized alternatives certainly pokes at there being some
> imperfections to the notion of ACID.

Surly the point is that at the logical level a 'concurrent action' should be *exactly* like a serialised action.

> It will take a good deal more
> than that to satisfactorily justify it being a good idea to drop ACID
> altogether.

BTW If (hopefully after reading some of the previous threads) you think there is some fundamental reason why we need transactions as the logical level, I'd be interested to here them.

Regards
Paul Vernon
Business Intelligence, IBM Global Services Received on Mon Sep 01 2003 - 20:52:07 CEST