Re: To DM or not to DM...that is the question?

In article <935163492.19487.0.nnrp-02.9e984b29_at_news.demon.co.uk>,   "Jonathan Lewis" <jonathan_at_jlcomp.demon.co.uk> wrote:
>
> The Jonathan Lewis (simplified) theory of Data Warehouse and Data
Marts -
> There's no such thing.
>
> There are only rather large databases which are physically rigged to
handle
> queries that need to thrash through large volumes of data as
effectively
> as possible. Anything goes structurally so long as it works in the
> situation
> you have to face.
>
> The Jonathan Lewis (simplified) theory of Multi-dimensional databases
> There's no such thing.
>
> A location in N-dimensional space is accessed via an N-tuple.
> An N+M column table (in any relational system) is the physical
> representation of an N+M-tuple.
>
> An N+M column table may therefore be considered the physical
> representation of an N-dimensional space where every point
> specified has M interesting facts.
>
> 'Multi-dimensional database' is a marketing term for a table
> with N co-ordinate columns and M fact columns. You can
> model this very effectively with Oracle 8i (amongst others)
> with a single table and N bitmapped indices.
>
> A more serious answer to your question -
>
> It is certainly the case the TNF is not a good structure for
> the sort of data trawling you have in mind. Protoype a
> simple non-updatable data extract process that simplifies
> an interesting bit of the database into a small number of
> simple tables. Call this a data-mart to satisfy anyone
> who feels the need for a buzz-word. Find out if people
> use this datamart, and work upwards from there, with
> the assumption that you might want to redesign the
> code every few months.
>
> When it really big, call it a data-warehouse; at this
> point you may understand the data and usage
> sufficiently to produce a fast-refresh, and historical
> trail extract. rather than a complete refresh extract.
>
> --
>
> Jonathan Lewis
> Yet another Oracle-related web site: http://www.jlcomp.demon.co.uk
>
> Vikas Agnihotri wrote in message <7pjq8j$s1a$1_at_nnrp1.deja.com>...
> >
> >My company (division, really) is just embarking on its "data
warehouse"
> >project. We have a rather complex OLTP system spanning the globe
using
> >Oracle, DCE, workflow management and a host of other
> >sophisticated technologies. It automates most of the business
processes
> >involved in the transaction, flag errors, print customer
correspondence,
> >interface with the General Ledger, Accounting systems, etc. Very
> >complicated.
> >
> >Is DM the right approach here? If not, how do we glean information
from
> >the monstrous OLTP system? Just saying "Use the right tools" is too
> >naive.
> >

I attended a Data Warehousing presentation by a Multi-dimensional database vendor and a consulting firm a few months back and they stated that the current thinking by the leading authorities in the field today is that you build a warehouse from your OLTP and other current business systems. This warehouse does not look like your OLTP but just holds the data you want to be able to track and do analysis from.

Then using this warehouse as a source you populate data marts, OLAP applications, and your multi-dimensional databases on a periodic basis. You then use these systems to generate your business analysis and information management.

This actually seems logical to me. I would also think that EMC and other disk vendors would love this approach.

To Vikas I would recommend that if hiring some experienced consultants to offer advice and maybe some training is not in the budget that you look for a specific limited need and try to develop for it as a learning experience which I think is similiar to, if not the same, as Jonathan's approach.
--
Mark D. Powell -- The only advice that counts is the advice that  you follow so follow your own advice --

Sent via Deja.com http://www.deja.com/
Share what you know. Learn what you don't. Received on Fri Aug 20 1999 - 12:50:28 CDT