RE: Re: Relational databases

I concur with Bradley's statements.  

 

RDBMS are challenged to handle all of the "dimensions" needed for any
architecture, because of all of the types or "relations" that are
simultaneously required for any give object (e.g.,
inheritance/classification/taxonomic relations, and association relations
(container/component, predecessor/successor, prior/current,
current/subsequent, alias, variant, reference, etc.).  Architectures have an
inherent "object" model, and RDBMS don't excel as object databases.  I use a
superset of the OpenGroup's object metaschema as my root object model for my
architecture and management approach.

 

Also note that the primary standard RDF store, that is, a metadata
repository, is the OMG MOF in all its various commercial and open-source
implementations.  The most advanced form of RDF is now specified by the Web
Ontology Language (OWL).  I professionally recommend OWL-Full front ends
(using RDF triples for text input and DirectedLabeledGraphs (DLG) for
diagrammatic input) and MOF backends/stores, with some
workflow/process-management mechanism in the middle.

 

As a tool to provide this, I often recommend the Agilense.com WebModeler.

 

 

Roy Roebuck

Ph: 703-598-2351

[email protected]

one-world-is.com/beam

Calendar: http://one-world-is.com/freebusy/roebuckr.htm

FreeBusy: http://one-world-is.com/freebusy/roebuckr.vfb

-----Original Message-----
From: Bradley P. Allen [mailto:[email protected]] 
Sent: Thursday, February 26, 2004 11:12 AM
To: [email protected]
Subject: [facetedclassification] Re: Relational databases

 

Andrew- While I would agree with you that reusability of information
architectural components is crucial for their use throughout an
enterprise, I believe it is a mistake to think that RDBMSs are the
only practical choice for sharing them.

Information architectural components such as taxonomies tend to be
trees or networks, and encoding these data structures in relational
tables involves compromises with respect to transparency and
maintainability. What typically happens is that the implementation of
such an approach ultimately has to present an API that wraps the
underlying relational encoding. This argues against the claim that
integration is "instant", since the developer has to either learn how
to code to the API or understand how to pack and unpack network data
structures in and out of the tables into their own application.

As Matt pointed out, scalable implementation of faceted navigation
requires in-memory representation of the information architecture.
Approaches based exclusively on executing queries against traditional
relational storage are one to two orders of magnitude slower,
certainly in our experience at Siderean and I believe in Endeca's as
well. Query optimization of the sort typically performed in RDBMSs
does not work well with the sorts of queries this application
requires. If for productivity's sake developers are coding to an API,
why not use more scalable storage underneath? It is true that most
enterprises have RDBMSs at hand, but they also have a lot of other
types of servers in their infrastructure as well: LDAP servers, HTTP
servers, POP servers. I think Matt and I would agree that information
architecture requires a new class of server, for the above performance
and productivity reasons.

I would further argue (and this is the major theme of my upcoming IA
Summit talk) that XML-based approaches for representing information
architecture, such as RDF and XFML, support reusability and sharing of
information architectures in the manner you are advocating, without
commitment to a particular storage approach, and in a way that
encourages reuse across enterprises as well as within them. -
regards, BPA


--- In [email protected], "Andrew D. Schriever"
<aschriever@p...> wrote:
> Matt and Claudio:
> 
> Thanks for the information -- very useful!  Regarding Matt's
question about the use of relational databases -- I'd like to share
some thoughts with the group and would be really interested in any
comments.
> 
> Our view is that the collection of taxonomies one might build to
support faceted search ought to be available for much broader use
within the enterprise.  If your search tool is useful then those
taxonomies must contain information of fundamental importance to the
enterprise.  More generally, we think that the starting point for
information architecture should be a basic corporate knowledge model
that can be applied to faceted search, business intelligence, product
catalog management, project and resource management, and a host of
other knowledge-driven applications.  If that's the goal, the only
practical choice for the knowledge repository is the RDBMS, since
virtually every enterprise of any size uses an RDBMS as its central
data store.
> 
> The joins can get intense -- you're right!  The flip side, however,
is that the RDBMS offers us the most sophisticated query optimization
engine ever developed.  Careful design can reduce the intensity of the
joins (though it can't be eliminated).  And integration with other
activities within the enterprise is instant, since most enterprise
data is present within the RDBMS, most applications access the RDBMS,
and most programmers have the skills to use the RDBMS.
> 
> It seems to us that commercial products tend to provide for
knowledge modeling or taxonomy building only to the extent that they
support a specific application.  In other words, document search
systems allow one to build taxonomies that support document search;
product catalog systems incorporate taxonomies that support the
catalog; content management systems often represent navigation with a
taxonomy; business intelligence systems have their own representation
of applicable hierarchies.  This duplication of knowledge models adds
unnecessary cost, complexity, and confusion and reduces the ROI
delivered by IA and knowledge modeling endeavors .
> 
> Does anybody out there have experience with attempts to provide
reusable, general purpose knowledge modeling as a base component of
information architecture?  Or with the use of the RDBMS for this
purpose?  Any success (or failure!) stories I ought to know about? 
I'd really appreciate your input.  Feel free to write me directly
(aschriever AT parkstreetsoftware DOT com) if you'd rather not hit the
whole group.
> 
> Thanks again!
> 
> Andy Schriever
> 
> 
> 
> 
>   ----- Original Message ----- 
>   From: Matt Eichner 
>   To: [email protected] 
>   Sent: Monday, February 23, 2004 10:38 AM
>   Subject: RE: [facetedclassification] Relational databases
> 
> 
>   Andy,
> 
>   You may also find some of the work going on at Endeca quite
interesting.  We
>   have developed a library interface that uses a faceted
interpretation of the
>   MARC record (not unlike the FAST project at OCLC) and enables
search fused
>   with this navigation.  As part of that, we read in the LOC
classification
>   schedule (we considered Dewey; swapping is a few days' work), in
addition to
>   several other navigation constructs (e.g., geography =
651$a/600's$z, etc.)
>   So, a user can execute a search on, say, "baseball" and then
navigate by LOC
>   class within the search results, and then cross that LOC class
navigation
>   with a particular geography (for instance, Japan).
> 
>   The main reason I write - we considered a database approach, but
found it
>   expensive/challenging to accommodate requirements for
dimensionality,
>   real-time navigation, and certain important features like fusing
navigation
>   with search (the joins became too intense, among other problems).
The
>   search, in particular, was important - it seemed that users wanted
this
>   familiar tool to at least take a first crack at reducing the
universe of
>   results.  If there is interest within this group, I can send out
some
>   snapshots, or arrange a better/live description (the snapshots are
~700K in
>   PowerPoint, so I don't want to crunch anyone's mailbox).
> 
>   Cheers,
> 
>   Matt.
> 
>   -----Original Message-----
>   From: Claudio Gnoli [mailto:gnoli@a...]
>   Sent: Monday, February 23, 2004 10:22 AM
>   To: [email protected]
>   Subject: [facetedclassification] Relational databases
> 
> 
>   """ [Andy Schriever : Jan 8]
>   one can apply the power of the RDBMS query engine 
>   to knowledge representation techniques such as semantic 
>   networks, class and object structures, and hierarchical 
>   classifications.
>   """
> 
>   Andy, you may find relevant for your purposes the work by 
>   View-based systems <http://www.view-based-systems.com>, 
>   where Steven Pollitt and Amanda Tinker have developed
>   several tools and applications for library catalogues and other 
>   databases, with special reference to the Dewey Decimal
>   Classification.
> 
>   Best whishes.
> 
> 
> 
>   ----------
>   Thanks for playing. 
>   Yahoo! Groups Links
> 
> 
> 
> 
> 
> 
>   ----------
>   Thanks for playing. 
> 
> 
>         Yahoo! Groups Sponsor 
>               ADVERTISEMENT
>              
>        
>        
> 
> 
>
--------------------------------------------------
----------------------------
>   Yahoo! Groups Links
> 
>     a.. To visit your group on the web, go to:
>     http://groups.yahoo.com/group/facetedclassification/
>       
>     b.. To unsubscribe from this group, send an email to:
>     [email protected]
>       
>     c.. Your use of Yahoo! Groups is subject to the Yahoo! Terms of
Service. 
> 
> 
> 
> [Non-text portions of this message have been removed]



----------
Thanks for playing. 




  _____  

Yahoo! Groups Links

*         To visit your group on the web, go to:
http://groups.yahoo.com/group/facetedclassification/
  

*         To unsubscribe from this group, send an email to:
[email protected]
<mailto:[email protected]?subject=Unsubscrib
e> 
  

*         Your use of Yahoo! Groups is subject to the Yahoo! Terms of
<http://docs.yahoo.com/info/terms/>  Service. 



[Non-text portions of this message have been removed]



----------
Thanks for playing. 
Yahoo! Groups Links

<*> To visit your group on the web, go to:
     http://groups.yahoo.com/group/facetedclassification/

<*> To unsubscribe from this group, send an email to:
     [email protected]

<*> Your use of Yahoo! Groups is subject to:
     http://docs.yahoo.com/info/terms/