|
Commentary: Synchronicity |
You can contact Nigel Pendse, the author of this section, by e-mail on NigelP@olapreport.com if you have any comments, observations or user experiences to add. Last updated on December 9, 2005.
The autumn of 2005 saw major releases from four of the major BI vendors: BusinessObjects XI Release 2, Cognos 8, Hyperion System 9 and Microsoft SQL Server 2005. Each of these was a major release which had been under development for years, and represents a significant investment by the vendors. Each of the vendors might have expected a significant competitive advantage in return for this investment, as well as a big market response to their expensive worldwide launches, but with all four releasing almost simultaneously, the impact is almost cancelled out. In effect, they’ve all had to run fast just to stand still.
Apart from the synchronicity of the four releases, they also share a common technical theme: responding to years of frustration with the poor integration of multidimensional and relational reporting, all four vendors have made big steps forward. In addition, IBM and Oracle have also shown signs of moving towards the same goal of a near seamless interface between the technologies. There probably will always be some differences behind the scenes, but business users aren’t interested in the differences, and don’t want to have to learn two parallel ways of performing ad hoc queries, analyzing data and formatting reports, based on arcane differences in data storage techniques. It’s even worse if both methods are needed as part of the same activity, with OLAP tools being used when looking at aggregated data, and relational tools when reporting on the underlying details.
What is intriguing is how all these vendors are heading for the same laudable goal from very different starting points, and using quite different methods. Despite the hype, none has yet reached the goal, and even if some eventually do, the implementations and even the motivations are quite different.
Traditional reporting vendors have always focused mainly on relational reporting, with OLAP being treated at best as a non-integrated add-on. In Business Objects’ case, this was particularly ironic as all BusinessObjects reports come via its own microcubes anyway. Bizarrely, for nine years, Business Objects has had a much better and more complete interface between its internal microcubes and relational databases than between microcubes and OLAP server cubes. We have regularly criticized Business Objects for not extending its universes to become OLAP-aware.
Finally, in XI Release 2, BusinessObjects universes are OLAP-aware, and can
initiate their own OLAP server queries, rather than having to have the microcubes
manually pre-populated from much more capable OLAP server cubes. This also means
that the same version of WebIntelligence can access both relational and OLAP
sources at the same time, and in the same report.
But the multidimensional query capability in XI Release 2, welcome as it is,
is still very immature compared to its relational equivalent, and may prove
disappointing in practice. For example, it has clumsy ways of handling parent/child
dimensions, non-additive aggregations, calculated measures, server-side ranking
etc. For these reasons, it cannot yet be regarded as a preferred way of performing
OLAP reporting, but at least to the extent that it can do so, there is now one
single WebIntelligence for both relational and multidimensional sources. Perhaps
proving the point that Business Objects still has some way to go on this journey,
it still has a separate OLAP Intelligence tool, based on the old Crystal Analysis
product that it acquired two years ago.
Cognos came from the opposite direction, as OLAP had until recently been more important to it than relational reporting. But Microsoft’s entry into the OLAP server market reduced the prospects for PowerPlay, so Cognos moved its focus to relational reporting, with the very successful launch of ReportNet in autumn 2003.
With Cognos 8, Cognos’ other BI products were moved to the new platform, though the financial applications lag behind. Now with three ‘Studios’ for building reports, analyses and queries, and each able to work with both pure relational, simple ROLAP and full MOLAP servers, Cognos has done a good job of blurring the distinctions between these three types of data source, though without making any changes to them. For example, PowerPlay itself remains the pure MOLAP that it has always been. Cognos, being more au fait with OLAP than Business Objects, has also made a more serious effort to support the advanced capabilities of OLAP servers, though it doesn’t go as far as the specialist third party tools available for Microsoft Analysis Services.
Hyperion’s challenge was slightly different. Much of its growth has come through acquisitions, and it is possible to spot features in System 9 that originated in at least 11 different companies. Hyperion’s primary goal with System 9 was not so much to blur the distinctions between multidimensional and relational reporting as between all its different tools and applications. System 9 is therefore more about delivering a common look and feel, common administrative tools and more consistent pricing to this collection of products than about deep level technical integration.
Deeper integration is expected at a later stage, but even this first stage sees some convergence of the former Brio Intelligence, SQR, Hyperion Reports and Analyzer products.
Microsoft’s approach is bolder than the others. Rather than converging the front-end tools while leaving the back-ends unchanged, or attempting to persuade the world to use SQL to query OLAP cubes, as Oracle has failed to do, Microsoft is attempting to meet the needs of both relational and multidimensional reporting from a multidimensional database.
Its proposition is that while transaction processing is best done in a relational database, business reporting that involves any form of aggregation (and most does) is best served from aggregate-aware OLAP (and usually MOLAP) cubes. It is setting out to demonstrate that the superior query and aggregation performance of MOLAP means that it can often provide more ‘real-time’ reporting than relational queries off the original source. To do this, it has changed the dimensional model in Analysis Services, which can now handle hundreds of attributes and dimensions in a single cube. The snag with this approach is that queries must be expressed in MDX, not SQL. This means that Microsoft is confronting an even bigger challenge than Oracle: it wants MDX to become the primary business reporting language, while Oracle was merely trying to get OLAP users to use the already very widely used SQL. And, of course, for this to work, the front-end tools must be optimized for Analysis Services 2005, and not just any old OLAP server that supports MDX queries.
If Microsoft can pull this off, it really will make OLAP and relational seamless, as the same server will deliver both through the same API and one set of tools. In theory at least, all queries will be delivered at MOLAP speeds, with no need to worry about the underlying data structures. When near ‘real-time’ is needed, the cubes can be configured to refresh frequently — perhaps every few seconds — though this will hurt the query performance, but no restructuring will be needed. So Microsoft must not only prove that this all-new technology works in realistically sized, real-world applications, but convince front-end tools vendors and users that this is the best way to do business reporting. The prize is well worth winning from Microsoft’s point of view, as success would inflict serious damage on competing relational database vendors like Oracle and IBM, and not just the much-reduced number of OLAP server vendors.
Even if Microsoft does not pull off this amazing feat, it is not taking any chances with its mainstream OLAP business: Analysis Services 2005 is a very impressive (if rather complex) OLAP server, which is almost certain to continue gaining OLAP market share. At the same time, SQL Server 2005 also contains much-enhanced reporting and ETL components, and Office 12 will have improved BI client tool capabilities (though these are unlikely to be as functional as the tools from BI specialists).
IBM is also moving towards MDX, but much more slowly. With the acquisition of Alphablox, it now has an MDX engine, which is being built into DB2 Cube Views. Currently, the engine is only used inside Alphablox, but in due course, it is likely to be extended and made accessible through an external API, such as XML for Analysis. This would allow DB2 to handle both SQL and MDX queries from the same database server, and deliver the performance gains from Cube Views-optimized materialized query tables. In some ways, this is the converse of what the BI vendors are doing: IBM will have one database and one server, but will be able to support both relational and multidimensional queries, presumably from different relational and multidimensional query tools (ie, one back-end, but multiple front-ends). The BI vendors work with separate relational and multidimensional sources, but attempt to provide a consistent front-end (ie, multiple back-ends, but one front-end).
Oracle is unique as it’s the only vendor still denying the need for a multidimensional query language, even though Express had always been strong in this area. Oracle is betting on the majority of its customers favoring the familiarity of SQL as a query language, even if this severely compromises the richness of queries. A version of the old Express command language is available through PL/SQL, but (so far) Oracle has been alone in ignoring the fact that MDX has become the industry standard multidimensional language, supported by all the front-end tools vendors, as well as server vendors like Applix, Hyperion, SAP, SAS, etc.
This resistance by Oracle seems to have done much more to harm Oracle’s BI revenues and BI market share rather than to MDX. If Oracle does succeed — against all the odds — all business information will be accessed through Oracle’s extended SQL and Java APIs, with some structured multidimensionally internally, but still stored in the same Oracle RDBMS.
In the end, the market will decide. In the short-term, the BI tools vendors seem to be responding best to the fact that important data is stored in both relational and multidimensional databases, but which the users would like to access seamlessly. In the long run, the database vendors will have a shot at actually converging the data sources into a single database, with Microsoft favoring MDX as the query language for business reporting while also offering SQL; Oracle offers only SQL; and IBM is now apparently backing both MDX and SQL.
This page is part of the free content of The OLAP Report, but ten times more information is available only to subscribers, including reviews of dozens of products, case studies and in-depth analyses. You can register for access to a preview of some of the subscriber-only material in The OLAP Report or subscribe on-line.
|
All information copyright ©2005, Business Application Research Center, all rights reserved