Chris Webb on "If You Think OLAP is Dead – Think Again" webinar
Yesterday I presented a webinar for the nice people at Pyramid Analytics, the first of two. I was paid for it of course (I have to earn a living somehow, you know), but the main reason I agreed to do it was because they wanted me to discuss interesting Microsoft BI questions rather than just demo their product – which is, as I’ve said here before, the most effective way of doing tech marketing in my opinion. In any case Pyramid Analytics is one of a small number of third-party tools I’m happy to recommend to my customers even without being paid.
Today’s discussion was on the topic of whether OLAP is dead and you can watch the recording here: https://www.brighttalk.com/webcast/13401/178899 [free sign-up required]
You won’t be surprised to find out that I don’t think OLAP is dead. By “OLAP” I mean the idea of a centralised model containing not just all your data but also things like how your tables should be joined, how measures aggregate up, advanced calculations and KPIs and so on. Using this definition both SSAS Multidimensional and SSAS Tabular are OLAP; I would say both pass the FASMI test for sure. The opposite of OLAP is of course the recently-fashionable self-service BI approach of tools like Power Pivot where every end user can create their own model, which is certainly more flexible but also less easily maintainable, leads to duplication of effort and data, and can be a problem for less technically-able end users. Both a single centralised OLAP model and a distributed self-service approach are useful in different circumstances and the latter certainly doesn’t replace the former. One of the reasons why I’m such a fan of Microsoft’s new direction with Power BI is that it recognises this fact.
The other big question that came up was what the relative strengths and weaknesses of SSAS Multidimensional and Tabular are. Before the webinar I ran a Twitter poll asking people whether they had used Multidimensional or Tabular on the last SSAS project that had worked on; from 151 responses, 48% said Tabular and 52% said Multidimensional. This even split doesn’t surprise me to be honest, based on my own experiences: both flavours of SSAS have different strengths and weaknesses and choosing between them can be very difficult. I’m sure new features like bi-directional cross-filtering and the improved development experience in SSAS Tabular 2016 will increase Tabular’s popularity in the future. Personally, I think if Microsoft were able to offer equivalent functionality to:
- Calculated members on non-measures dimensions
- MDX SCOPE statements
- Parent/child hierarchies (without the performance penalties)
…in SSAS Tabular then there would be very few good reasons to choose SSAS Multidimensional for new projects.