Re: What is integrated?

From: Paul Vernon <paul.vernon_at_ukk.ibmm.comm>
Date: Mon, 11 Nov 2002 22:44:11 -0000
Message-ID: <aqqko0$13l0$1_at_sp15at20.hursley.ibm.com>


"Scotty" <invallid_at_invalid.spam> wrote in message news:tversusqgv74c4oa219emcvsi1n1orilt4_at_4ax.com... [snip]

> Christ it sounds like a load of gibberish when I re-read it.

You said it! But it's not much worse than a lot else that is written on this ill founded subject.

Some pointers:

 1-jan-2002 and 01-01-02 are not different domains. They are different representations of the same value from a particular domain (that I tend to call TimePoint - i.e. a TimeDuration from some fixed point in time). This doesn't mean that your data sources will recognise such logical differences however.

Logically, and you are asking on c.d.theory here, you should correctly model your data sources first (noting things like domains, domain representations, domain definitions, constraints, attributes and relations), then normalise them (including adding then normalising temporal attributes if history is required to be kept) and only then integrate them.

> These ETL tools are often responsible for the success of an ODS project

Do you know that from first hand experience, from some closely reasoned argument, or are you just guessing (or worse, taking somebody else's word for it)?

Maybe you should try to rigorously define what 'problem' an ETL is attempting to 'solve'.

Regards
Paul Vernon
Business Intelligence, IBM Global Services Received on Mon Nov 11 2002 - 23:44:11 CET

Original text of this message