From a feature-by-feature perspective, the latest releases of UDB and O8i are very competitive BI platforms. But although on the surface any differences might seem trivial in determining overall value, a comparison raises several distinguishing factors. For our purposes here, I'll break down the areas of comparison into five categories: infrastructure, online analytic processing (OLAP), spatial data, supporting technology, and vendor strategy.

For each category, I'll focus on two aspects. First, I will identify the high-level features that you must consider and compare between the products. (This process will satisfy the traditional feature list assessment.) Second, I will examine the underlying strategic issues that influence the implementation of the category in general and the features involved specifically. My goal is to give you not only a comparison of critical features, but also an understanding of how those features are implemented and their influence on your organization.

Infrastructure

Infrastructure features are those that buyers expect of any database engine that can scale to meet the volume and performance requirements of medium to large warehouse environments. UDB and O8i both provide the parallelism and partitioning features of leading products and address the same infrastructure issues. (See Table 1.)

Infrastructure Features	Specific Features	DB2 UDB 7.x	Oracle8i
Parallelism	Bulk Loader	Yes	Yes
	Queries	Yes	Yes
	Indexing	Yes	Yes
Partitioning	Tables	Yes	Yes
	Index	Yes	Yes
Management Tools		Yes	Yes
Extraction,		Yes -- DB2	Yes -- Oracle
Transformation, and Loading		Warehouse	Warehouse
Java		Yes	Yes
XML		XMLYes -- DB2 XML Extender	Yes -- Context Cartridge
In-Memory		Yes	No
TABLE 1 Infrastructure Capabilities.

Now for the differences. In UDB, users can use DB2 Control Center to manage privileges, performance, licenses, and the warehouse databases. In addition, DB2 Warehouse Manager, a separate product, provides supplementary warehouse management features such as the ability to tune predictive queries and track resources and queries. As for Oracle, O8i provides extensive tools that help DBAs manage a warehouse database, including expert systems for performance and space management.

Both UDB and O8i offer toolsets for extraction, transformation, and loading (ETL). In UDB, standard ETL functionality is provided by default with DB2 Control Center. For an extended feature set, you can purchase DB2 Warehouse Manager to manage ETL across distributed environments. Oracle's Warehouse Builder represents its competing ETL tools. The toolset includes many of the features of the original base products, Carleton Corp.'s Pure View family, which were leading, mature ETL tools when Oracle acquired the company in 1999.

Support for languages such as Java and extensible markup language (XML) also enhance BI because the Internet or intranet will be the principal venues for publishing and accessing analytic data. As expected, UDB and O8i are replete with integration points between these two languages. UDB fully supports all published standards for Java, SQLJ, and JDBC. O8i supports many of the standards and includes features such as an integrated Java Virtual Machine. With DB2 XML Extender, UDB can support document storage, search, and retrieval as well as de- and recomposition of XML documents into and from tables. Oracle's Context Cartridge extends the functionality of O8i to include advanced text search on any XML object; it offers the ability to parse, store, and search any XML document within its RDBMS.

Table 1 also includes an entry for in-memory database (IMDB), which can speed Web searches by a factor of 10. IBM made the feature available with its latest release of UDB. Oracle, although it currently does not have an IMDB at the time of this writing, is expected to include it in a future release.

It's no surprise that both IBM and Oracle provide similar functionality and feature sets in this category. We could debate how well each product implements parallelism or the granularity between Java implementations. But at the end of the day, these products have similar infrastructure features.

Data Marts and OLAP

Warehousing is replete with indexing challenges: not only size and volume, but also the unique nature of data marts built using the star schema. With this type of data structure, the notion of merely having bitmapped or B-tree indexing is insufficient. Star schemas, by design, mean that a WHERE clause constraint is typically associated with dimensions used to perform the aggregations of the fact table. Traditional index techniques result in a Cartesian graph of all the dimensional constraints; consequently, the optimizer attempts to scan through an inordinate number of fact table rows.

In order to maintain performance, database vendors have implemented indexing strategies often referred to as star joins. In this approach, optimizers access data stored in star schemas using a phased approach. As a result, users can perform "speed-of-thought" analysis, or at the very least, avoid getting constantly bogged down by poorly performing queries. Both UDB and O8i have tuned their respective optimizers to deal with star schemas and have explicitly implemented a star join option. (See Table 2.)

Data Mart Features	DB2 UDB 7.x	Oracle 8i
Star Joins	Yes	Yes
ROLAP	Yes ­ Native SQL implements	Yes ­ ROLAP functions available
	the OLAP addendum for SQL '99	through Oracle Express
Aggregation	Yes ­ Automatic summary tables	No -- Not declarative
Aggregate Navigation	Yes	Yes
OLAP	Yes -- DB2 OLAP Server	Yes -- Oracle Express
TABLE 2 Data mart and OLAP capabilities.

Other data mart capabilities worth mentioning are aggregation and aggregate navigation. The nature of multidimensional analysis dictates the extensive use of aggregation. For example, a user may want to see sales summarized by region, month, or product class. If the granularity of your data mart is store, day, and product ID, then your query may have to summarize millions of records to deliver the correct aggregated values. You could aggregate "on the fly," but only slowly, and the problem could be exacerbated if the query is a common one. Thus, in this case, you may decide to preaggregate fact data against a particular dimension hierarchy or combination of dimensions. The result of this preaggregation is then stored in a permanent, aggregated table.

Having the ability to define aggregation points and the associated rules that can run automatically as the data mart is refreshed is a standard feature of UDB. Of course, not having the feature doesn't mean you cannot use your ETL tool or hard-coded transformation logic to achieve the same objective. When you have aggregated tables, you then face the challenge of navigating data access paths to use the right aggregated tables in order to maximize performance. Therefore, aggregate navigation is an important area that benefits from a planned, vendor implemented solution. Both UDB and O8i provide a database-resident means for navigating user queries to the optimum aggregated tables.

But data marts go beyond star schemas. For example, although relational OLAP (ROLAP) implementations are rare, ROLAP is an important concept that customers should expect their vendors to address. OLAP cubes cannot scale much beyond 100GB in size: If you need to perform multidimensional analysis on 1TB of data, ROLAP is the answer.

Both UDB and O8i provide ROLAP functionality. In the case of UDB, ROLAP is implemented through standards established to native SQL as well as through its implementation of DB2 OLAP Server. Having ROLAP as part of the extended SQL, however, gives you the advantage of not having to use its OLAP product to do ROLAP. In the case of O8i, ROLAP functionality is achieved through its OLAP product, Oracle Express.

The final major data mart feature in our comparison is the provisioning of OLAP. Major RDBMS vendors recognize the advantage of integrating an OLAP toolset into their databases, and all of them have purchased, or partnered with, a company to provide OLAP technology.

Oracle purchased one of the oldest OLAP engines, Express, in 1995. Express is a full-featured multidimensional database. Unfortunately, Oracle has an unsatisfactory track record with the product. To start, Oracle has done a poor job of integrating Express with its RDBMS. For years, Express lacked the seamless integration required between OLAP and the RDBMS, which is expected these days. Moreover, Oracle did not consistently invest resources to keep pace with product features as the OLAP segment matured. For this reason, current market research from The OLAP Report (olapreport.com) suggests that Express is losing significant market share and will continue to do so.

Even before Microsoft set the precedent of bundling OLAP with the RDBMS in SQL Server, Oracle was giving away Express in large, bundled license deals; as a result, many Express licenses are associated with Oracle Sales Analyzer or Oracle Financial Analyzer.

This is an unfortunate position for Oracle. If any company has the means to press an OLAP/RDBMS agenda, Oracle does. Instead, its OLAP strategy is littered with self-inflicted problems. Through missteps and missed opportunities, Oracle finds itself competing against better product offerings. DB2 OLAP Server, IBM's DB2 OLAP product, is one of them.

DB2 OLAP Server is based on the integration of the best-selling OLAP engine, Hyperion Essbase, with DB2 UDB. This integration has matured over the last year; IBM has even announced DB2 OLAP support for OS/390. Moreover, in Microsoft fashion, there is now a DB2 OLAP Starter Kit included in the purchase of UDB. The kit is a limited license, meaning that IBM is seeding future sales as opposed to simply losing all potential revenue for its OLAP offering. Thus, where Oracle finds its market share quickly eroding, IBM finds itself in a growth trend for OLAP.

An effective integration of OLAP and the RDBMS has long-term implications beyond immediate product sales. With OLAP quickly becoming a commodity and an expected feature of the DBMS, it changes how OLAP will be offered in the near future. I expect that OLAP application vendors will move away from developing their own technology and instead license or simply exploit multidimensional tool sets from leading RDBMS players. The same holds true for the ROLAP niche. As major RDBMS vendors implement more ROLAP features into the engine itself, why purchase ROLAP functionality from non-RDBMS vendors such as MicroStrategy Inc.?

Spatial Data

No overwhelming user demand yet exists for geographic information systems (GIS) data and related spatial technology, but vendors are quietly building them into their products as standard features. This phenomenon reminds me of land development in the suburbs: As you drive through an undeveloped area, you may notice giant bulldozers leveling land, and then forget about it. A few months later, you take that same route and see that business and residential communities have popped up overnight, creating the illusion of instant growth. Of course, there was nothing "instant" or "overnight" about it. The expansion was planned in advance, and you just didn't happen to know.

GIS will hit the broader market in this manner. Spatial data types, specialized indexing, and extended SQL must blend into the RDBMS before the "killer apps" that spellbound users can emerge.

Thus, to remain competitive, leading RDBMS vendors must address the issue of spatial objects, and both IBM and Oracle have done so. Each vendor has an integrated, kernel-level, spatial solution. (See Table 3.) The principal difference between these vendors is in how they implement spatial data: DB2 UDB Spatial Extender integrates spatial data into the kernel of the database engine as a spatial data type, GEOMETRY. Moreover, UDB has extended the SQL in accordance to the Open GIS standards. This approach ensures that all operations against the spatial data are done exclusively through SQL functions.

Spatial Data Features	DB2 UDB 7.x	Oracle 8i
General	Yes -- DB2 UDB Spatial Extender	Yes -- Oracle Spatial Cartridge
Indexing	Index technology specific to handle spatial data types. Indexing is required.	Index technology specific to handle BLOBs. You are not required to index the data.
Storage	Stored as GEOMETRY data type with compressio. Replication of the data is available.	Stored as a BLOB with limited compression. Oracle parallel Server cannot replicate objects.
TABLE 3 Spatial data capabilities.

Oracle has taken a different track with its Spatial Cartridge. It has elected to store spatial data as a binary large object (BLOB). Performing operations against the spatial data, users have access to a combination of extended SQL and C functions, known as "packages." The particular solution, although effective, is not compliant with established standards. In this sense, the Spatial Cartridge is somewhere between an application server and an extended kernel.

Ultimately, the adherence to established standards ensures future interoperability. For IBM users, the adherence to established standards for such a complex technology is great news. It lets them exploit the full marketplace when choosing supplemental products for data access and spatial analysis.

In contrast, anything short of adherence to standards binds you to a particular technology suite or guarantees future integration problems. In the case of O8i, third-party products will have to accommodate Oracle's custom C functions. This process may require an extra layer of middleware.

Supporting Technology

For BI applications, supplemental products can enhance the infrastructure needed to manage, exploit, and publish information. Two such product offerings come to mind: portal technology and data mining tools. With the exception of CASE technology, both IBM and Oracle appear to address these technologies with comparable offerings. (See Table 4) However, their implementation strategies are substantially different.

Supporting Technology	Specific Features	DB2 UDB 7.x	Oracle 8i
Portal	General	Yes — IBM Enterprise Information Portal provides federated search capability for both structured and unstructured data. Also included is text-mining capability to provide classification, categorization, and Web crawling. EIP integrates workflow management for the portal infrastructure through IBM's MQ Series.	Yes — Oracle Portal provides full access to any structured information within the Oracle RDBMS or any federated data source defined to Oracle.
	Unstructured Data	Yes — IBM Content Manager provides management of unstructured data powered by DB2.	Yes — Oracle Portal can access a variety of data sources including disk files and Web sites. It can also manage files within the RDBMS through the Internet File System(IFS).
	Federated Queries	Yes — IBM EIP	Yes — Resident Capabilities
Data Mining		IBM Intelligent Miner for Data provides data mining with alorithms for advanced mining functions including: clustering, associtaion, and prediction. IM supports mining directly against DB2 tables or flat files.	Oracle Darwin is a data mining software suite providing parallel implementation of classification, regression trees, neural nets, and so on. The product supports VLDB mining.
CASE Tools		No — DB2 UDB provides metadata interoperability between the Data Warehouse Center and ERwin.	Yes — Oracle 8i is integrated with the Oracle Discover 2000 CASE tool.
TABLE 4 Supporting technology.

Infrastructure Components	IBM	Oracle
Operating Systems	Full Offering	Partners
Middleware	Full Offering	Full Offering
Database	Full Offering	Full Offering
Data Warehouse	Full Offering	Full Offering
Management	Full Offering	Partial Offering
BI	Full Offering	Full Offering
Portal	Full Offering	Full Offering
Content	Full Offering	Partial Offering
ERP	Partners	Full Offering
CRM	Partners	Full Offering
TABLE 5 The big picture: How infrastructure reflects company strategy.

For example, IBM's Enterprise Information Portal (EIP) is designed and marketed as a standalone application, the goal being for portal vendors to use it as rapid development and deployment platform for portal applications regardless of whether UDB is involved. This approach is dramatically different from Oracle's; that company's portal technology is specifically tuned to support the development and deployment of Oracle-based Web sites. Thus, Oracle follows the conventional strategy of providing technology to enhance its own product offering, while IBM is offering a standalone tool set that can serve as the underlying technology for any Web site, even those running on Oracle.

Vendor Strategies

Table 5 shows a published Dataquest comparison between IBM and Oracle infrastructure components. The published findings cross the spectrum of software segments from each company.

This list illustrates the agendas driving both these companies. For example, according to this research, IBM has more depth in management, operating systems, and content software than Oracle. On the other hand, Oracle has an overwhelming lead in applications software such as enterprise resource planning (ERP) and customer relationship management (CRM). That suggests a fundamental difference in strategies.

To press this argument, we can find other evidence of divergent core strategies in recent partner announcements. For example, in May 1999, IBM and Siebel Systems Inc. announced that the Siebel Front Office application suite will support DB2 UDB as the preferred DBMS. In December 1999, SAP and IBM announced that SAP will now switch its internal systems running on AIX, Solaris, and Linux to UDB, indicating for the first time that SAP favors DB2 over Oracle for some Unix platforms. Finally, in June 2000, PeopleSoft and IBM announced that PeopleSoft will use UDB as the company's primary Unix development platform for PeopleSoft8.

In their respective areas, these three leading providers of application software built much of their early success on Oracle platforms. However, as Oracle continues to demonstrate an unrelenting drive to become an application shop, these same vendors found themselves anxious to reduce their dependence on Oracle.

Both UDB and O8i are well positioned to keep, and perhaps grow, future market share. However, their philosophically different goals resonate across their respective product offerings. In essence, Oracle wants to become the ultimate ERP vendor. The company is perhaps the leading RDBMS vendor and has the underlying technology and, therefore the means, to readily integrate functionality such as OLAP, spatial data, and portals into its ERP packages. No surpise here: since releasing Oracle Financials, Oracle has steadily evolved from a traditional RDBMS vendor into an application powerhouse. With its introduction of CRM and e-business services, Oracle has only solidified that reputation.

IBM's agenda is markedly different from that of Oracle. IBM wants to become the de facto standard technology foundation that all application companies employ, and this philosophy shows in its ever-growing ranks of leading application partners. Where Oracle makes application partners nervous, those same partners find an ally in IBM. And given that most feature-by-feature comparisons are a wash, IBM will continue to build a reputation of being the preferred partner of choice.

What's Your Choice?

What are the most important criteria for deciding the value UDB or O8i have as BI platforms in your organization? It comes down to a simple question: Are you an Oracle shop? If you are, the answer is easy: Go with Oracle-based solutions. Similarly, if you are planning to become or evolve into a homogenous shop based on Oracle technology or applications, you should continue to invest in Oracle. It remains a leading provider and arguably offers the widest selection of enterprise products. On the other hand, if you are building an adaptive infrastructure or best-of-breed environment with heterogeneous applications and technology, then IBM should be your first choice.

Rate This Article Rating: 1 (best) 2 3 4 5 (worst) Comments: Optional e-mail address:

Michael L. Gonzales (mlg@starfocus.com), a database developer for more than a decade, manages The Focus Group Ltd., a consulting firm specializing in ROLAP and OLAP techniques and technologies. He has also written several books and conducts data warehouse seminars across the country.