DMR or Relational Model

[sukanya]

Hi,

I have more than 14 dimensions and 8 million fact. Which type of FM model should I build? Relational or DMR. Are functions used in DMR similarto cube if I have build using DMR. what should be the criteria that I need to consider. Kindly explain

Thanks & Regards,
Sukanya

[MFGF]

Hi,

I have more than 14 dimensions and 8 million fact. Which type of FM model should I build? Relational or DMR. Are functions used in DMR similarto cube if I have build using DMR. what should be the criteria that I need to consider. Kindly explain

Thanks & Regards,
Sukanya

Hi,

The choice between a dimensional model and a relational model isn't something a third party can dictate. What influences your choice is the business requirements you need the reporting solution to deliver. Do your users need to drill down to more granular levels of detail on an ad-hoc basis? Do you need to deliver relative time analysis? Is your data flat or does it naturally lend itself to hierarchies? Are you delivering static scheduled reports or do users need self-service analysis capabilities including slice-and-dice? Etc etc. If you have 8 million fact rows and a need for a dimensional model, DMR may or may not provide the performance you require - it has to aggregate all the numbers on-the-fly. Consider DMR as being a "virtual cube" - the basic capabilities it has are very similar to using a real cube.

MF.

[sukanya]

Thank You Meep. I have a composite key (6 different columns) Join between Dimension and Fact in our star schema. To create business key in DMR FM model, do I need to have business key defined in the underlined database table that joins to the Fact? Since, business key in DMR doesn't allow to use multiple columns.

[bus_pass_man]

That's this question isn't it?

http://www.cognoise.com/index.php/topic,31367.0.html

http://www.cognoise.com/index.php/topic,24030.0.html

4. Double posting. Please do not post the same question in multiple threads. If you want a topic moved to a new thread, please just ask a moderator - we're happy to help. Ideally you will post your topic in a new thread anyway, and not need this service. Posting the same question in multiple places can be confusing for people looking for an answer later on, and can also mean that people spend time answering questions that have already been resolved elsewhere.

[sukanya]

If I am creating a DMR, do I need to build the crossreport and prompts using dimensional functions? Or is it ok to build like a relational report. what I am trading off by using relational way of building the reports?

Kind Regards,

[MFGF]

If I am creating a DMR, do I need to build the crossreport and prompts using dimensional functions? Or is it ok to build like a relational report. what I am trading off by using relational way of building the reports?

Kind Regards,

If you're using a relational model, use relational reporting techniques. If you're using a dimensional model (including DMR) use dimensional reporting techniques. There are many reasons why using relational techniques with dimensional sources is bad - you can get incorrect, inconsistent results. One common mistake is to use detail filters. Take a look at #4 in the FAQ post below for an explanation:

http://www.cognoise.com/index.php/topic,27563.0.html

MF.