Cognos OLAP: Understanding performance differences
This weekend, I was reading a very interesting discussion on LinkedIn, on the topic of OLAP (specifically Cognos OLAP). On this thread, ex-Cognoid Norman Bobo does an excellent job of describing the differences between Cognos PowerCubes and TM1.
In his comments, Norman describes in detail several differences between the two technologies. Here is a summary of his key points:
1) Purposes – PowerCubes are a read-only BI solution – TM1 is read/write
2) Dates – PowerCubes have the concept of date dimensions – TM1 does not
3) Levels – PowerCubes have the concept of dimension levels – TM1 does not
4) Attributes – TM1 has strong support for attributes – PowerCubes do not
5) Cube building – very different (TM1 is an in-memory solution, PowerCubes are not)
6) Data Scalability – PowerCubes do not manage large dimensions very well
7) User Scalability – TM1 is not so easily scaled
8) Cognos BI Integration – as a legacy Cognos technology, PowerCubes are more tightly integrated
(I strongly encourage you read the entire thread here)
From a User Experience perspective, I believe IBM would position their various offerings as specialized OLAP approaches to optimize for specific types of applications:
In truth, each solution does has it’s own unique strengths and weaknesses:
It is also interesting to consider how each solution approaches roll ups and aggregation. Specifically:
PowerCube – data structures are pre-aggregated:
TM1 OLAP – sources aggregate on the fly:
Cubing Services – optimizes the most common paths or caches as you go:
So, which is best for Read-only BI? Well, of course it depends on the customer requirements. As Norman points out in his article, data volumes and user volumes are key considerations. Some general guidelines are as follows:
It is certainly important to be aware of the frequency & type of reporting required by the user community, as well as the size and geographic location of the user community (e.g. centralised or decentralised). It also worth noting that Cognos internal testing clearly demonstrated that Powercubes scaled very well with increasing concurrent user volumes, whereas TM1 cubes did not.
Other points of note:
Key takeaways….
Footnote:
Interestingly, many years ago I actually built a neural network model (using Cognos 4Thought) of PowerCube build statistics from Cognos customers around the world, in order to identify the factors which had the biggest impact on Cube build times.
Transformer Cube Build Factors
The neural network model clearly demonstrated the strong correlation between cube build times and the number of categories in the PowerCube.
