Intelligent Enterprise

Better Insight for Business Decisions

RSS
Webcasts
Whitepapers
Subscribe
Home

March 1, 2003

API Wars

Web services platforms — .Net and J2EE — take the fight to the analytic arena

by Seth Grimes

Application programming interfaces (APIs) are more than simple communications channels for interconnected software components: Their scope and limits often determine the data structures, methods, and performance characteristics of interoperating programs. Adoption or neglect of particular APIs may therefore mean the success or failure of larger computing-platform initiatives.

I presented a first look at emerging analytic APIs — XML for Analysis (XMLA) and Java Online Analytic Processing (JOLAP) — in my previous column, "XML for Analysis Decoded." Version 1.0 of the XMLA specification had been released in April 2001 by an industry consortium led by Hyperion Solutions Corp. and Microsoft — SAS signed on as a third principal sponsor a year later — and JOLAP development was proceeding under Java Community Process (JCP) rules with an expert group led by Hyperion and Oracle representatives.

More than 20 months later, neither API is ready for general use. But both have evolved, as has the overall Web services story. What I've learned studying the development of XMLA and JOLAP and related APIs for data mining and metadata management illuminates that story.

XML for Analysis

It was apparent from the start (although not openly admitted) that XMLA is a Microsoft-dominated initiative driven by Microsoft's needs and timetable. Sanjay Poonen, vice president of worldwide marketing at business intelligence (BI) vendor Informatica Corp., told me, "Microsoft is trying to control [BI] front ends the way [it] attacked the Cognos and Hyperion servers."

Microsoft CEO Steve Ballmer has "bet the company" on XML as the basis of the company's .Net Web services initiative and as an interchange format. The reality: Microsoft understood the urgency of escaping the strictures of the Windows Component Object Model (COM). COM lets users embed Excel charts in Word documents but is wholly unsuited — in terms of performance, scalability, and security — to the demands of distributed Internet computing.

The initial XMLA specifications — v.1.1 was released by the XMLA Council in November 2002 — retain the Multidimensional Expressions (MDX) discover and execute methods from Microsoft's OLE DB for OLAP (ODO) Windows API. XMLA replaces ODO's Object Linking and Embedding transport with Simple Object Access Protocol and provides an XML schema for marking up dimensional-model elements.

The XMLA Council is adapting MDX for the Web services environment as mdXML, which is scheduled for release as part of XMLA 2.0 in late 2003. In a conference call a few months back, Council members told me that establishing a "standard linguistics interface" for OLAP, namely MDX/mdXML, is a key goal. They liken their work to the establishment of SQL 15 years ago.

Microsoft's Tom Conlon said during that call that XMLA 2.0 will be the primary API for the next version of Microsoft Analysis Services, code named Yukon. Yukon is, clearly not coincidentally, slated for release in the same time frame as XMLA 2.0, although Microsoft executives are now talking about the first half of 2004.

The 18-month gap between the release of XMLA versions 1.0 and 1.1 surely reflects slower-than-expected Web services evolution and adoption. The 1.1 release followed an XMLA Council meeting focusing on interoperability testing: ensuring that XMLA clients and service providers from different vendors can communicate. A meeting attendee told me that the testing had been largely successful, with more than half of the tested client/server pairs fully interoperable by the end of the exercise. And in a post-meeting conversation, he confirmed my take on the initiative: "Look at the reality of how the whole thing has been driven: It's Microsoft, Microsoft, Microsoft." A follow-up XMLA Council meeting is scheduled for April 2003.

The J2EE Platform

Oracle is, by choice, odd company out, with no intention of supporting XMLA or mdXML. Oracle similarly doesn't support XMLA's ancestors, Microsoft Analysis Services and MDX, offering instead SQL, Java, and other OLAP interfaces. Oracle OLAP product development and management executives Greg Dorman and Bud Endress are quick to denigrate XMLA as not controlled by a standards board and unlikely to be widely adopted, and they see their Java-oriented strategy as more credible. In fact, I've found that various vendors plan to provide XMLA servers but not to XMLA-enable their client applications.

Oracle prefers Java standards and sees easy migration (from their existing Java interface) to JOLAP.

A Hyperion employee, John Poole, is the JOLAP specification lead. He and I spent a couple of hours discussing JOLAP and Java interfaces for metadata packages that are compliant with the Object Management Group's (OMG) Common Warehouse Metamodel (CWM), a framework for building metadata models tailored for types of applications including OLAP and data mining and for various data models including record, relational, and multidimensional. Hyperion is leading efforts in both camps, XMLA and Java 2 Enterprise Edition (J2EE), but while Poole's colleague Matt Abrams handles the XMLA side of things, Poole clearly prefers the J2EE platform: "Perhaps one of the most significant aspects of JOLAP is that it is grounded in a number of established standards for interoperable systems such as CWM and the Java Community Process's Java Metadata Interface (JMI) and J2EE Connector Architecture. JOLAP represents a confluence of several important industry standards targeted toward ... advanced analytics."

SAS also plays a leading role in both efforts. One of its strategic objectives, being realized in v.9 of the SAS system, is to create an analytic computing platform that interoperates with any significant external package. Its goal, as I see it, is simultaneously to stay relevant to statisticians and broaden its software's appeal to the mainstream market.

Data Mining in the Mix

The family of JCP analytic standards includes a data-mining API, JDM. In this instance, the expert-group lead is an Oracle engineer, Mark Hornick. Mark wrote me: "The real value of data mining occurs once models are deployed in applications. JDM is an important step toward enabling 'production data mining,'" and "Expert group members, including IBM, SAP, SAS, Hyperion, and Oracle, have been quite supportive of the standard from its inception." The expert group targets mid-2003 for release of the final specification.