Re: The original version

I received an email from one of the Anchor Modeling authors in relation to this thread. With this message I would like to say that I do not wish to engage in a private correspondence about this thread. If someone wishes to express his/her opinion related to the thread, let them do so in this user group.

It seems that certain things in this thread are not clear, and so I wish to briefly clarify.

1. In my paper, I do not use 6NF. I do not see a theoretical significance in 6NF. The authors of 6NF have not given a procedure or solution that puts a relvar into 6NF. If someone wishes to put a relvar into irreducibility form, he must do so completely on his own with the careful analysis of all possible cases, neither less nor more.

I do not see a reason why the authors of Anchor Modeling are thrilled by 6NF. Again, I argue 6NF does not give a solution for putting a relvar into 6NF. The authors only define what they call 6NF. On the other hand, the main problems in Anchor Modeling are solved by my procedure named (a), which I presented in this thread on May 26th (the first message in the thread).

2. My response to Bob was determined by a small impreciseness expressed in the following sentence: “Let us consider a relation with five attributes that are mutually independent”. 6NF is about JDs, not about FDs.

I have forgotten to say that in this case we have an all-relation key and that because of that 6NF does not make sense in this case, because (for example) the key can be composed of 20 attributes. Actually, I intended to say this when I said that attributes are mutually independent.

> There is one other thing here, which is more important, which is badly
> done in the "Anchor Modeling. This is about how to do the transition
> from E / R model in the relational model and vice versa. I think it is
> necessary to define the mapping from E / R to RM, then the inverse
> mapping for the given mapping and in the end it is necessary to define
> the composition mapping. In my model I have at the outset, the binary
> concepts. Each binary structure has its own unique identifier of the
> state. Therefore, each tuple or a binary concept is uniquely defined.
> In "Anchor Modeling" They start from the E / R and go in the RM, so do
> 6NF, and return to the E / R. But it was not discussed in the paper,
> so it's not clear how to do it. For example the definition of 6NF is
> interesting:
> (A table is in 6NF iff it satisfies no nontrivial join dependencies at
> all.)
> We can note that mapping of schemas between two db models can be
> complex, for examples it can include constrains.

3. Recently in this user group problems related to a design using two data models were discussed. Vadim posted some text about this problem. In anchor modeling the authors use the following two data models, ER and RM. In the paper they often use the term table. However, in their paper they do not explain how the two data models are worked with, which is a crucial question. In ER, FDs and JDs are not the same way as they are in RM, so we cannot work in ER (six) normal forms. In RM there is no entity or relationship at all. There are many papers by R. Fagin about schema mappings. Microsoft has also published papers about this topic (see Phil Bernstein, Melnik, Alagic).
The conceptual schema is an especially serious problem because it has components which are at a very high level of abstraction (a concept for example). I have tried to solve the schema (inverse) mapping here by introducing a definition of concepts, abstract objects (m-objects in my paper) and other terms.

Vladimir Odrljin Received on Tue Oct 05 2010 - 01:32:22 CEST