Re: Identity modelling (was: dbdebunk 'Quote of Week' comment)

mAsterdam wrote:
> Alexandr Savinov wrote:
> [snip]
> > In general I think that we lack information on "identtity modeling"
> > althoug it is as important as data modeling itself. Identity modeling is
> > a separate topic, a dual part for data modeling. In other words, we can
> > model identity ignoring object properties.

Would this be what I have been simply calling entity modeling, but where I leave out all weak entities and non-key, non-foreign-key properties? Or are you suggesting modeling the ID's rather than the entities with their id's?

> > And it may well be rather
> > complex model. It will involve entities without properties - only
> > identities. The following properties of identity make this task rather
> > difficult:
> >
> > - Identity is distributed among many entites.

Hmmm. Several entities can all use the same attributes as unique identifiers, but I'm not sure what it means that identity is distributed among many entities.

> > It can be hierarchical. For example, an element of categorization might have several segments
> > each specifying relative position. A fully qualified identifier then is
> > composed of several identifiers (for example, several primary keys taken
> > from different tables - having one primary key is not enough).

Can you give an example?

> > - Identity cannot be considered without its scope. For example, a
> > physical address is retricted by the scope of one computer, a primary
> > key might be restricted by one database etc.
> >
> > - Logical/physical is a relative characterization rather than absolute.

agreed.

> > Memory handle is really physical for an application program that uses
> > it, but it is logical for operating system w.r.t. to absolute offsets
> > in physical memory (offset may change while memory handle does not
> > change). In this sense all those disputes about lgoical/physical are
> > meaningless without specifying the context.

or at least without a common understanding of the context

> > Primary key may well be
> > viewed as a physical identifier from the point of view of some higher
> > level identification mechanism, say, global id. This means that global
> > id is permanent while primary key it substitues may change.
> >
> > - Any identifier is based on some environemnt that provides a coordinate
> > system that it uses to produce its own identifiers. In other words, any
> > new identification system is based on some lower level identification
> > system (environemtn or context) with its scope and structure.
> >
> > Currently no one model provides anything that would deal with the topic
> > of identity modeling.

it might if you don't try to extract these identities from their entities. If you model your strong entities and their identifiers, what falls through the cracks routinely if you use OO modeling with composition, aggregation, inheritance, etc?

> > We can model columns and object properties more or
> > less successfully. But another side is still in darkness. We have tools
> > to implement identification mechanisms ourselves but we do not have a
> > theory for that.

interesting.

>
> This deserves - at least - a thread of it's own.
> If it would be a /. moderation I'ld have a hard time to choose between
> 'Insightful' and 'Interesting'.

agreed. --dawn

> No time to comment at this time, though :-(
Received on Tue Aug 30 2005 - 06:02:15 CEST