Re: Polygon data modelling question

From: Qingqing <zhouqingqing_at_excite.com>
Date: 17 Jan 2002 19:34:31 -0800
Message-ID: <cee59be7.0201171934.4ff22767_at_posting.google.com>


paul_geoffrey_brown_at_yahoo.com (Paul G. Brown) wrote in message news:<57da7b56.0201170857.4ed017db_at_posting.google.com>...
> "John Palmer" <jopalmer_at_mail.vt.edu> wrote in message news:<a24amu$qd$1_at_solaris.cc.vt.edu>...
>
> Starting in the early 1990s, many DBMS products began to support the idea
> of DBMS 'extensibility'. The idea is that a concept like a 'polygon' (or a
> path, or a graph, or a rational number, or a complex number, or a Java class
> corresponding to a mailing address) are modelled within the database as a
> SQL data type (think relational domain (Hi Chris!)). All of that class/type/
> domain's behavior is implemented by linking the appropriate shared library
> (or jar file) into the DBMS and havings its functions invoked by the DBMS
> in response to SQL queries issued by the application. For example:
>
> ...
>
> So in a nutshell, the most efficient way to address your requirements, in
> terms of operational efficiency (a good DBMS will have indexing to
> support these extensions), data model flexibility (no more will you have
> need to convert complex real world objects into a lower level set of
> char arrays and numbers) and developer productivity, is to make use of this
> kind of technique.
>
> But if you'd like, you can still just shove the data into a blob and
> write all of this stuff outside the DBMS in a middleware layer.
>
> Hope this helps!
>
> KR
>
> Pb

Hi Paul, what you mentioned here actually is an old and important and unsolved problem in database area ( and also many other computer fields). The basic problem can be called "k-NN query" or "range query". And in database the indexing is called multi-dimensional indexing which try to settle this problem.

In some commercial database, they support R tree, which is presented at the beginning of 80s. In fact, in 1999, Roger Weber gived a simple but efficient method called "VA-file" showing good performance... There is a very long story and tons of research on this ... IMO, the basic conclusion is: (1) R tree or other R-tree improvements only work well in <= 6 dimensional space; (2) VA-file is a promising method for high-dimensional space ...

If you want to know more about this problem, just use google search "Roger Weber" then go to his website and find out a lot of stuffs on this problem. However, the practical suggestion is: don't sink into this problem if you just doing programming, coz it is really hard. If you want to implement an indexing method for your application, try VA-file method.

regards,

Qingqing Received on Fri Jan 18 2002 - 04:34:31 CET

Original text of this message