ORDBMSs: Universal Databases

Software developers used to design and optimize relational database management systems (RDBMSs) to support OnLine Transaction Processing (OLTP) systems. Vendors then added configuration parameters to let us use OLTP systems for decision support systems (DSSs), client/server, and data-warehousing applications. Systems must now also manage nonstructured data--everything from documents and Web pages to graphics and audio clips. Document management systems, object-relational database management systems (ORDBMSs), and standalone Web site programs might offer good proprietary solutions, but many organizations prefer that their RDBMSs handle the data in a centralized fashion.

Universal servers (sometimes called ORDBMSs or hybrid systems) need to deliver a synthesis of relational and object database technologies, offering more flexible data-type support and a better fit with Internet architecture and distributed objects. Universal servers have been available only since late 1996, so their eventual role in database management is unclear. Each of the five major RDBMS vendors uses a different architecture for its universal servers.

IBM.IBM has supported object-relational data types, as well as content-specific search engines that extend SQL, since June 1996. IBM's technology relies on extenders rather than on cartridges (Oracle), DataBlades (Informix), or a middle layer (Sybase). Extenders take advantage of user-defined types (UDTs) and user-defined functions (UDFs) to extend SQL to handle nontraditional data. For example, you can use UDFs, which are registered directly with the Universal Database (UDB) kernel, in any SQL statement. DB2 UDB continues to rely on a pointer- and metadata-based approach to extended data types. Several of IBM's partners, including MapInfo and ESRI, are developing additional extenders.

Informix.Informix released INFORMIX-Universal Server (IUS) in late 1996 as a combination of classic INFORMIX-OnLine Dynamic Server and Illustra in a tightly integrated single server that natively stored object data types. This architecture was based on DataBlade engine extensions. Competitors such as Oracle asserted that bad DataBlades could crash or otherwise corrupt a database. This problem was unlikely, because Informix planned to certify DataBlades and give users options for running DataBlades on the client. In late 1997, Informix announced a more conservative approach to universal server technology that relies on Informix's flagship OnLine Dynamic Server, renamed Dynamic Server. Object extensions (via DataBlade modules) and Informix's online analytical processing (OLAP) server (MetaCube) are now add-ons to Informix Dynamic Server, which is available on multiple platforms, including Windows NT.

Microsoft.Microsoft outlined its universal database server strategy at its May 1997 Scalability Day. Microsoft's SQL Server 7.0 (code-named Sphinx) will offer universal server support through a Windows-centric Object Linking and Embedding-Database (OLE-DB) approach in which external active servers store complex objects. Currently, SQL Server 6.5 supports binary large objects (BLOBs), but it doesn't offer SQL extensions to handle queries for audio, image, or video content. Rather than further extend SQL Server's proprietary Transact-SQL (T-SQL), Microsoft favors a more client-centric approach to optimize OLE servers' features. SQL Server, which is part of BackOffice, is available on only NT.

Oracle.Oracle released its universal server, Oracle8, in 1997. Although Oracle8 is less object-oriented than Oracle planned, it provides a conservative migration path for thousands of companies that use Oracle servers. Oracle8 uses cartridges to support spatial data and text searching, as well as other extended data types such as image and time series. Oracle and third parties offer cartridge extenders, or programmers can write their own.

Sybase.Last April, Sybase announced its three-tiered Adaptive Server architecture; Adaptive Server 11.5 shipped late last year. Adaptive Server has a middleware layer, the Component Integration Layer (CIL), rather than Oracle's cartridge or Informix's DataBlade approach. Adaptive Server supports additional object data types via Component Data Stores (CDSs), which Sybase and third-party vendors offer to support spatial data types such as image store, image search, text or document search, and time series.