Post PASS thoughts on BISM

Reposted from Thomas Ivarsson blog with the author's permission.

I think that most of you are aware of the announcements during PASS last week when Microsoft presented their new software offerings in the BI area. The reactions was mixed from different great BI bloggers like Chris Webb,  Marco Russo and Teo Lachev and more will come.

BISM(Business Intelligence Semantic Model) is the next server based, stand alone PowerPivot, that will not require SharePoint 2010 server. At this moment it is not totally clear what will happen but let us say that it is the column based relational engine in the Analysis Service sphere that will only talk serverbased DAX with the clients and that you know from PowerPivot. BISM will not have a full understanding of MDX so todays scripts in SSAS will not translate into DAX. It is more of a side by side relation with some limited mutual understanding.

During the next years we will have both the ‘old’ Analysis Services multidimensional engine,SSAS, side by side with the BISM/VertiPaq relation and columns based in memory engine.  SSAS will have features that will take BISM some releases to catch up with that is mostly about complex calculations in MDX combined with dimensions(calculated members in dimensions, script assignements and so on). SSAS will still be here for many years and hopefully it will get improvements during releases.

From an IT-technical perspective this can be bad news but I will focus more on end user requirements. In the best of worlds the multidimensional storage in SSAS could be directly translated to the relational in memory model of BISM and MDX calculations could be translated to DAX but this is not the case.  A decision must be made of what kind of data marts that will benefit most from being in a multidimensional model and a column based, relational and in memory model. This is how I see it.

SSAS has both attributes and dimensions in the so called Unified Dimensional Model or UDM. This model offers great navigational help for end users with natural hierarchy navigation from top levels to leaf levels in parts of the hierarchy. On the other hand, if you want to relate different dimension this is done by nesting the hierarchies on rows or columns in a grid. In SSAS/UDM combining single attributes, and not hierarchies, from different dimensions, is not the strongest feature of that server. It is slow.

Some end users like the top down analysis approach with the birds eye perspective on data but other end users prefer a more flexible approach by combining single attributes from different dimensions and start with analyzing dimension data on leaf levels.

This last group is the one that I recognize daily as being less satisfied with the current versions of SSAS and much of their work would start with flattening SSAS cubes down to leaf level nested dimension levels. I assume that this group of users can be helped with both the current version of PowerPivot and the release of BISM in the next version of SQL Server, DENALI.

BISM and PowerPivot are mostly marketed as being fast with massive amounts of data but I think that the flexible attribute based model is also important.