Re: Modelling objects with variable number of properties in an RDBMS

Bernard Peek wrote:

> >> An EAV database could be used to collect
> >> metadata during the early development stages, and that metadata could
> >> then be used to build the final database system, replacing the temporary
> >> EAV model.
> >
> >Please elaborate on this, because I can't make head nor tail of it.
>
> In the early stages of agile development you probably have only a vague
> idea of what entities and attributes are going to be required in the
> final system.

This is an odd statement. How can you even define an entity if you have no clue what attributes it consists of ? Besides, it's been known how to handle the entity evolution for, like, ages. One can use schema evolution DLL statements (ALTER TABLE), inheritance (see Tutorial D), etc.

> By building a system where the users can add entities and
> attributes without involving the developer you allow the users to define
> what the final system needs.

So what you get is an unstructered pile of string of characters split into three subpiles. There is no notion of relationship too.

To sum up briefly: you have no logical model, no relationships, no constraints, poor performance due to the need for excessive joins and character stings conversions, no data types. What's the gain ? With this approach, you do not need a relational database. Just three indexed flat files would suffice for this kind of stuff.

>After collecting the metadata you have the
> information required for the data modeller to put together a properly
> structured relational data model.

What exactly do you mean by "metadata" ?

"to put a properly structured relational data model", is it not what database design modelling is about ? Why not just do it with a pencil a piece of paper instead of wasting time on "EAV" ?

>
>
>
> --
> Bernard Peek
> London, UK. DBA, Manager, Trainer & Author.
Received on Mon Oct 31 2005 - 20:37:58 CET