Hi Julie/All,

My apologies.

I paraphrased a 2 page article into half a sentence and a couple of people have noted I mis-represented the article.....which, when I re-read my comment, I certainly accept did...:-)

When I said 'transaction processing' I meant read transactions from near real time reporting or OLTP systems that need to read integrated data.

Not so much update tranasctions. ODSs usually have precious little 'updates' happening in them, though it is not unknown of course.

If you would prefer not to read my 'paraphrase' but read the article the link is below. This is what my comment was based on.

In his conclusion, Ralph sumamrised the article as follows:

"In this column I have made what I hope is a strong case for satisfying the new realtime requirement with specially constructed, but nevertheless familiar, extensions of existing fact tables. If you drop nearly all the indexes and aggregations on these special new tables and pin them in memory, you should be able to get the combined update and query performance that you need."

In the end, this article is putting the case for using dimensional modeling for similar processing requirements for which ODSs with a 3NF flavour have been used over the last 6-7 years. The 'real time'

partition.

Personally, I'd have been skeptical if I had read this article 2 years ago when it was first published. Getting dimensional models updated quickly enough to support what Ralph terms 'realtime partition' in a sizable environment like a bank or a telco is a reasonably tough job.

And 2.5 years ago, the ones I knew about were 'work in progress'. Good signs, but not 'production'.

However, having just implemented a large ODS on top of a dimensional model, whos purpose was data integration so that OLTP systems could read the integrated data, I agree that a dimensional model can be updated quickly enough to be made to support near real time 'read transaction processing' (I'll be specific this time...;-).

Also, in the project I reference we do plan to perform updates to the ODS by exposing the DataStage jobs that populate it via the new Ascential SOA. But that is work in progress....my one is not production yet, but we have spoken with a few other ASCL clients and we have built tests so we are very confident we will be able to make it work.

Best Regards

Peter Nolan

Data Warehousing Consultant

Mobile: +353 879 581 732

-----Original Message-----

From: owner-dwlist@datawarehousing.com

Sent: 09 August 2004 23:39

To: dwlist@datawarehousing.com

Subject: RE: dwlist: The compelling business case for an archival non-lossy

From: "Julie Kimball" <julie@ralphkimball.com>

Just for the record, Ralph has never written an article "discussing transaction processing on a dimensional model." I just asked him. ;-)

Thanks,

Julie