Re: Why is database integrity so impopular ?

From: <patrick61z_at_yahoo.com>
Date: Sat, 1 Nov 2008 07:18:44 -0700 (PDT)
Message-ID: <793e70b6-8d17-4b7f-9aaa-7e2759bae784_at_u18g2000pro.googlegroups.com>


On Nov 1, 8:57 am, "Walter Mitty" <wami..._at_verizon.net> wrote:
> <patrick..._at_yahoo.com> wrote in message
>
> news:fab13a41-6c86-45fc-9cf2-88c74e9c338a_at_i20g2000prf.googlegroups.com...
>
> > I think that relational database theory is too limitting for some
> > applications. I believe that the modern database needs to be split
> > into component parts so that not everyone has to be saddled with the
> > relational part.
>
> I think you might have meant "the relational data model" where you wrote
> "relational database theory". True?
>
> Could you give an example?
>
> If you design part of your database to be relational, and another part to be
> something else that's not conformant to relational, how do you use data
> from the two parts together in an integrated fashion? Or is that too much
> to ask?

Well in theory everything I want could be handled by a relational database and normalized tables. In practice, some situations are too expensive to be normalized for me as currently implemented. The ideal situation would be for my database product to offer fully normalized looking access to what are essentially repeating groups or cached direct access, ie., if I could have 'select whatever from wherever where key = someinteger' be actually and guaranteed translation to 'return whereever[someinteger * elementsize]' then sure, I'll just put up with a wordy dialect that translates to an array reference. Sometimes I want to be that precise in how many computer cycles I've decided to burn for that particular operation. I'm a convenience store clerk with rented webspace for my forum for instance. I'm spamming digg with my relational dbms using blog entry and I don't want my $10 a month server to be 'buried'.

Yes, I have this for ram, but can I set array elements 1, 3, and 5 to value 'something'' and have all three assignments either be persisted or not and no partial state? Can todays rdbms offer me this without even using a hash, just indexing into adjacent locations of storage? Could I do it with sql today, passing an out of band pragma that says when I'm describing a normalized table that replaces the repeating groups I loved so much, can I still tell it that "I'd really like this puppy to be at random access speed into contiguous storage localized in reference to the current left side of the join?" Received on Sat Nov 01 2008 - 15:18:44 CET

Original text of this message