Analysis Services 2005 introduced a brand new architecture, namely the UDM (Unified Dimensional Model) allowing a level of abstraction to be created between the Business Analyst and the underlying Data Sources. This meant that the inherent complexity of the Multi-Dimensional database access would always be transparent to the end user, regardless of the storage options selected by the development team. This is all well and good and much appreciated by all concerned. The key object in the UDM is called the Data Source View (DSV) which creates the unified viewpoint of the data irrespective of where or how it is stored. It also opens the door to Real-Time Analytics, something that was not possible before. But is this really a viable option?

Real -Analytics refers to the concept of being able to analyze business trends through a multi-dimensional data source (aka a "Cube") with the data being updated in real-time. The classic alternative previously was to accept the "latency" inherent in an OLAP solution. For instance, in the common "daily load" scenario, where incremental updates are loaded once a day, the natural latency would be 24 hours. Users would recognize that the data would be up to 24 hours old, but since we would be typically analyzing over many months or even years, this would not be a big deal. However, nowadays in industries such as Financial Services, time is money and real-time doubly so. (Ever read the Hitchhikers Guide to the Galaxy? I think I misquoted...). Many companies are now looking at this solution provided by Microsoft's Business Intelligence and the new UDM architecture. Imagine being able to analyze the financial market trends as they happen through visualization techniques and dashboard style applications. This is a new requirement that many companies are willing to throw some big resources towards.

However, the implication of this new solution is that as data is updated real-time in our operational systems, the OLAP system will be able to access that data directly with zero latency. No need for a Data Warehouse when we can get the data directly from the source. This will add a huge processing overhead to our already stretched OLTP systems. "Not on my watch" I hear you DBA's say, as you make a cross with your fingers and hiss like a cat. But if we are designing a new application with a new hardware budget maybe we can do some realistic capacity planning and justify the power that would make such a solution viable. Maybe. Or perhaps we don't need all of the data to be real time. Analysis Services allows us to create partitions in our cube and each partition can have its own storage options. i.e. some partitions can be real time and others not. One major brokerage client we worked with in Boston wanted Foreign Exchange data to be real-time but the rest of the cube data to be subject to the normal latency. This may be a viable solution as long as we perform full integration testing including performance benchmarks ahead of time. No surprises.

So next time Microsoft talks about Real-Time Analytics, you'll know what they're talking about - and when to show them the door...

Cheers Brian

Recent blog posts:

The International Dirty Word Database

Sliding Doors or Sliding Windows?

Database Design – build a blueprint for your database