Oracle Open World Day 3 - BI Suite EE & BI Publisher

Mark Rittman -

I woke up this morning with the mother of all hangovers after spending the evening in an Irish Bar around the corner from the Moscone, but at least my bag was still with me with my laptop still in it. Compared to leaving your laptop in a bar, sorting yourself out with a strong coffee and a fry-up is fairly straightforward stuff and I made it along to Matt Elumba's talk on BI Suite Enterprise Edition Architecture, at 10.45 in Moscone South.

About half of the presentation was a general recap on BI Suite EE and it's positioning, and we got to see what I presume to be the 10gR3 V10.1.3.2 "Maui" release in action. Here's a few screenshots taken with my camera:

The architecture part of the session walked through the basics of the BI Server and the BI Web Server, and there was a nice section on fragment handling (subsets of dimension tables used to assist lookups), multi-user development in the new release, the various levels of caching and the calculation wizard in the administration tool. From this slide it looks like the Enterprise Semantic Model is now called the Common Enterprise Information Model.

Straight after the BI EE Architecture session was Mike Donahoe's one on BI Publisher. Mike used to be a Discoverer PM and now he's moved over to XMLP, which is now being renamed "BI Publisher" with additional integration points into BI Suite Enterprise Edition, and BI Suite SE some point in the future. Some highlights of this session included a demonstration of the integration with Oracle BI Suite EE (you create the request, then go into the XMLP report browser and select "OracleBI" from the report source drop-down - presumably Discoverer will be another source in the drop-down when Discoverer integration is completed), the latest version of the Word add-in (a new Discoverer menu toolbar menu more options, new chart and table wizards), upload of report definitions from within Word and a slightly spruced-up look and feel for the Enteprise product.

Next up was Ed Suen's "Oracle BI Enterprise Edition Best Practices" - Ed was one of the co-founders of nQuire and now heads up development within Oracle BI. I was particularly looking forward to this session but at the end I had mixed feelings - it tried to cover too much and instead didn't cover anything in enough detail. I've fallen in to this trap myself with my OLAP Best Practices paper - you feel obliged to include every best practice you can think of, when in fact just five or six covered to a proper depth would be more useful - but it was still very valuable to hear Ed' take on the BI EE development process, and to go through metadata creation best practices and some common design patterns. Anyway, some of the key points that came out of the presentation included:

  • the key differentiator for BI Suite EE is it's model-driven approach, which helps minimize requirements suprises as you develop the application
  • BI EE handles large data volumes through function shipping to the back-end DW database or OLAP server, rather than creating it's internal data stores and trying to scale those
  • The model-driven approach makes early prototyping easier - prototyping rather than a long requirements-writing process is the preferred development method, with a fully-functional prototype featuring role-based dashboards being delivered at the end
  • A typical BI EE project would involve the prototyping/requirements gathering phase, a performance tuning phase, regression testing and subsequent projects/iteration
  • When you build the Common Enterprise Information Model, make sure detail-level transaction data is kept in separate subject areas to aggregate, summary data, as mixing the two creates unusable presentation catalogs and bad query performance
  • Although BI EE supports the integration of data across data sources, this shouldn't be used as a way of creating a virtual data warehouse (EII) - according to Ed this just doesn't work.
  • Understand the semantics of fact-based partitioning and logical table sources - by default every measure is a separate query block, but coalescing measure mappings in a single logical table source allows you to optimize by coalescing measures to reduce the number of queries
  • When implementing BI EE, resist the urge to rebuild the source data warehouse - instead create a handful of aggregates to speed up the slowest report, migrate these eventually to the source warehouses, use the caching and aggregate navigation features of EE to speed up the warehouse in-place.
  • When building analytics catalogs, keep to no more than seven top-level folders and seven columns per folder, minimize detailed columns (use navigation to access these instead), focus on adding measures and cross-fact calculations - this is where the value of EE is achieved, and use consistent order and naming - dimension folders at the top, measures at the bottom, and measure folders are named specifically so that end-users can distinguish between dims and measures.

Anyway, I'm off to the blogger event soon, so it's back to the hotel for a shower and then out again.