How will this change the BI OnDemand service? Well, any application combining lots of data with a need to do fast analysis will benefit.

To give you some idea, SAP has introduced Hana-based Workforce Planning and Smart Meter Management apps, and it's planning other stand-alone apps including Sales and Operations Planning, Intelligent Payment Brokering, and Cash and Liquidity Management. Sales & Operations Planning will let managers create and test the financial impact of various pricing, supply, and demand scenarios. Intelligent Payment Broker will let finance types assess invoices and quickly analyze the profitability or cost of various financing options. The Cash and Liquidity Management app will help managers predict cash flows so they can plan collections, risk management, and short-term borrowing. These and other apps are expected to roll out in 2012.

What all these apps have in common is the need to quickly look across many dimensions within large data sets. Having up-to-the-second data isn't crucial in all these examples, but it would be maddening if you had to wait minutes, or even more than, say, 15 seconds for each and every query or what-if scenario to run. Whether it's a purpose-built in-memory app or an equally challenging app running on Bi OnDemand, Hana's in-memory capabilities will ensures split-second results.

[ Want more on Oracle Exalytics? Read Oracle's Big Plans for Big Data Analysis. ]

Now about Oracle, when it comes to its claim of being able to run any BI or ERM application on an in-memory appliance, I say what's the point? This claim is based on the fact that the appliance will run Oracle Business Intelligence Foundation middleware. So, yes, technically it will run any BI app developed on that platform. But how many of those apps were developed with in-memory capacity and flexible multidimensional analysis in mind? Given that this capability didn't exist before, I'm guessing the answer is none or very few (and to be fair, the same is probably true of applications built to run on SAP's BI OnDemand). It could well be that some apps bogged down on slow servers and growing data sets will speed up on Exalytics, but at what cost? Oracle has yet to put a price on the machine.

Exalytics will also support a new in-memory version of Oracle Essbase, the OLAP database upon which most Oracle EPM apps were built. Clearly these are a better fit for the power of in-memory computing, but the existing apps were developed with the constraints of the old Essbase in mind--such as needing pre-defined cubes and queries. Will these apps suddenly be able to do ad hoc analysis against any dimension, or support unfettered scenario planning? We have yet to find out, but I suspect a bit of application revision will be in order, adding the kind of visual, data-exploration interface Oracle demonstrated at Open World. In fact, during a question-and-answer session at the event, a group of Oracle applications executives acknowledged that they'll be developing new apps specifically to exploit Exalytics. I won't be surprised if the list includes the same sort of apps on SAP's Hana to-do list.

Whether you're an SAP customer or an Oracle customer, keep in mind that Hana and Exalytics aren't always necessary. This gets back to the Ferrari analogy. Why drive a high-priced speed demon if a Fiat 500 will get you where you need to go? You don't need the power of Hana or Exalytics unless your application demands fast execution of complex analyses or what-if scenarios against high-scale multidimensional data. Lots of organizations appreciate the flexability of in-memory exploration even if they're not burdened with huge enterprise-grade data sets. That explains the success of products like IBM Cognos TM1, QlikView, Tibco Spotfire, and Tableau. And it also explains why SAP offers BusinessObjects Explorer, a multidimensional data-exploration tool that can be used with or without the power (and expense) of an in-memory appliance behind the scenes. Oracle's only option for avoiding pre-built cubes and aggregated data is Exalytics, a separate hardware investment.

The bottom line is that real, real-time isn't always achievable and the power of in-memory analysis isn't always required. So before you get swept up in the hype and spend big bucks, take a long, hard look at your top priorities and data-analysis needs.