The advent of "Universal Servers" is one of the most significant changes now rippling through the database technology market. However, it's not happening overnight, and there are clear signs that vendors are struggling with this transformation. Most customers seem to be taking a wait-and-see attitude, and I think that's sensible at this time. If building a Universal Server is hard, migrating is probably not going to be easy for many shops either.
First, I think it's interesting to examine the names and terms used to describe this technology. When I first researched this database trend about two years ago, the term "object/relational" was most common. Object/relational implies that the DBMS engine integrates both object-oriented and relational features, usually without a clear bias toward either architecture. In mid-1995, IBM Corp. used the term "Relational Extenders" to describe add-in datatypes for DB2. I thought that term was an accurate description of IBM's technology. Shortly after Informix Software Inc. acquired Illustra Information Technologies, both Informix and Oracle Corp. adopted the term "Universal Server." Recently, IBM came closer to Oracle's and Informix's terms with its new label "Universal Database." Clearly, "Universal" is more of a marketing term than a technical term. At best, it implies improved support for a broader range of datatypes. But I think it backs away from more ambitious goals to truly blend object orientation and relational features.
Names and terms are just words; the actual technical capabilities and limitations of the technology delivered to customers matter more than mere monikers. Here again, backtracking is evident. Oracle made many bold (and by now old) promises for object support within its DBMS. Oracle8 is finally moving from slides to software, but even Oracle admits it remains a work in progress. ("That feature will be in 8.1. . .") Good software is never really finished, but it's hard to argue today that Oracle has fulfilled its early claims regarding the degree of object support. Informix may be shipping its Universal Server, but even that product is not yet the complete superset of Informix's RDBMS and Illustra's server. (See Judy Davis's article on page 75 for details on current and future features in Universal Servers.) Computer Associates International Inc. (CA) aborted a project to build an object/relational DBMS. CA now promotes an object ODBMS (Jasmine) as a better alternative to blended object/relational hybrids.
When you clear away the hype, what remains is an effort by RDBMS vendors to improve support for complex datatypes, and in some cases complex relationships, within the framework of what is still largely an RDBMS. If you really need an object DBMS, using a pure ODBMS will probably be your best bet for the foreseeable future. (This month's cover story by Nelson King on page 58 takes stock of the state of ODBMSs.) I think the most noteworthy aspect of the Universal Server effort will be support for installable domains -- Informix DataBlades, Oracle Cartridges, IBM Relational Extenders, and so forth. This advance is valuable because, in the long run, it broadens the scope of the DBMS server without actually requiring a completely separate server to manage specialized datatypes. I think this will prove to be a reasonable middle-ground approach that lets customers manage and integrate more data more easily. Dozens of vendors (many of them relatively small) are building installable domains for specialized datatypes. I think this approach will become more common, if not widespread, over the next few years. For this to happen, customers must become comfortable with the idea of using third-party software in a database server.
Although I believe extending relational technology is an important advance, I also recognize that the uptake of this technology by customers is progressing slowly. A gradual adoption is to be expected because database servers are truly mission-critical -- a failure can halt business operations -- and the nature of this change involves substantial modifications to the core DBMS engine. Relational DBMSs were not overnight successes, either. It will take time for customers to conduct hands-on evaluations, deploy proof-of-concept projects, and accumulate enough confidence to apply Universal Servers throughout their organizations. It is also important to realize that relational DBMS servers will not go away, just as legacy database managers have not gone away. Over time, newer technologies become the primary solution for new systems; they usually supplement rather than replace what they follow. What's your view of Universal Servers? Let me know via email; my address is mfrank@mfi.com.