On Sun, Sep 27, 2009 at 2:22 PM, David Fetter <david@fetter.org> wrote:
> On Sun, Sep 27, 2009 at 08:26:27PM +0200, InterRob wrote:
>> Dear David, dear all,
>> I very well understand what you are saying...
>
> Clearly you do not. What you are proposing has been tried many, many
> times before, and universally fails.
I've been refraining from jumping on this due to time constraints, but
this statement is silly. We have a system that does almost exactly
what the OP wants although the implementation is slightly different:
we use an EAV like model with strong typing and build set / subset
forests to maintain arbitrary hierarchies of relationships. Our
reasons for doing this are similar to the OPs; it's for research (in
our case medical research). We maintain over 200,000 pieces of end
user generated metadata, describing what would be in a conventional
relational model over 20,000 columns and some 1,000s of tables but the
actual physical model is some 40 tables. Yes, the flip side is, such
a system won't support more than 1,000,000s of transactions per day,
but that's not why you build them.
>
> That your people are failing to get together and agree to a data model
> is not a reason for you to prop up their failure with a technological
> "fix" that you know from the outset can't be made to work.
>
Spoken like someone who has always had the luxury of working in areas
with well defined problem domains... I can't tell you the number of
people that told us exactly the same thing when we started on it.
That was 8 years ago. Not only can such systems be built, they can be
made to scale reasonably well. You do need to understand what you are
doing and why: the costs can be high, but when it comes to research,
the benefits can far outweigh the costs.
--
Peter Hunsberger