I found this presentation quite nice:
https://www.google.de/url?sa=t&rct=j&q= ... UP4F2W0mZw
Its not really a how-to but gives some good hints to make decisions about how to proceed.
The book "IBM Cognos TM1 - The Official Guide" (Oehler, Gruenes, Illacqua, 2012) gives some knowledge from p. 341 ff about scaling of TM1 models:
https://www.ibm.com/developerworks/comm ... 07?lang=en
Did you see this post already?
http://www.tm1forum.com/viewtopic.php?t=3774
Generally you might want to avoid dimensions going to deep into the data. I had a marketing requirement, which aimed to analyse checkout receipts for a retailer. They had facts described by outlet, employee id, cash desk id, articlenumber per itemline and a timestamp. They wanted me to make this data viewable via a dimension date and also a dimension time, which should contain not only hour, but
also minute and second of the timestamp. Obviously TM1 would never get a chance to perform only one single aggregation (summing >1 value having the same dimensional set of keys) as the combination of ids was absolutely unique. And the amount of data to load spanned over 12 years and covered 180 shops, quite huge.
This worked not well, but any proposal to reduce the degree of detail wasn't accepted. So we stopped that development. In advance a presales guy from IBM tried to find a solution, the ideas weren't taken by business, but for informational purposes here they are:
- Make a TI-loaded big cube holding only the fact data. It should have only flat dimensions (which are lists of granular elements, only level 0, no consolidations). And other cubes for the user access picking up the data from that big cube via rules (and maybe some filtering). Its also proposed in the above mentioned book (p. 342).
- Another idea was to build a less detailed cube and implement drill through paths: A user drills down to the most granular level of this cube and gets data which is aggregated, and doesn't have the degree of detail needed. But via a connected TI process the data, which makes up the aggregated cell value navigated to can be loaded just then without persisting it in the cube. So the cube serves as a n-dimensional tabel of contents and the click on the drill leads to the retrieved details. An example how to build this you find in this book: "IBM Cognos TM1 Cookbook" (Ankit Gard, Packt Publishing, Birmingham 2011, p. 215 ff.) https://books.google.de/books/about/IBM ... edir_esc=y
Hopefully your stakeholders are more open for alternatives, if you can't reduce the amount of data.
I hope you will find something that helps.