Andrew po-jung Ho wrote:
>
> Hi Wayne,
> Thanks for the information! I am going to make some comments below in order
> to make sure that I understand what Synapses/SynEx is about. Please let me
> know if my understanding is wrong or incomplete.
>
I am not going to claim that I am a Synapses expert at all, I pointed
out the project because it has had experience federating health records,
albeit at the systems level. I pointed that out and said I did not know
if federating at a smaller level (individual or collective of like
minded individuals) is something than Synapses can inform us about.
> >
> >The key thing to remember here is that Synapses and SynEx are attempting
> >to federate at the system level.
>
> Right. I (and Renner et al) think this is harder and not necessary. The
> system level "federation" will be incrementally achieved through the use of
> the less ambitious "Linkers"-metadata as these interchanges are needed.
>
I really don't think Renner, et. all was suggesting going lower than the
'system level'. When they talk about users wanting to maintain their
viewpoints, I believe they are talking about users of a system's
viewpoint, which they have selected because of it's viewpoint. All the
data in the paper are from the systems level.
"...our approach works as a layer on top of the existing DBMS
systems."
To avoid the non-scalable aspects of M to N transformations, Renner,
et. al. propose automatic translations and use of an intermediate
reference schema. They then give the requirements for achieving
adequate automatic translations. At some level of abstraction, however,
there must be commonality. Rather than talk meta-data (which is just
information about data) they talk about a process of schema
transformation thus:
source schema ==> reference schema (IDEF1X data model to be specific)
==> receiver schema.
They then generate, using a formal language and another common element,
a shared vocabulary, automated transformations into and out of IDEF1X
data models.
In Synapses the reference schema is the pre-CEN standard model, and the
main difference is that the mediators are hard coded, unlike Renner's
proposal. I don't see how to avoid the M to N problem without an
intermediate common representation and neither do Renner et. al.
