Databases: The object of your desire

The database has been with us for years and will be for many more ? but what will it look like in the future? Danny Bradbury finds out

The database has been the supporting strut of the mid-range software industry for years. Ever since computing became commercially viable, before Windows or even desktop clients as we know them existed, the database formed the centre of the computing landscape. Although this landscape hasn?t changed as quickly as the peripheral developments that sprang up around it, it has still metamorphosed over the years, and the tectonic plates of the database world are slowly shifting again.

When databases started out, they were flat-file affairs, with limited functionality. When hierarchical databases were introduced in the late 1970s and early 1980s, they were designed for very fast operations on localised records. The main problem with them was that they were not designed with diverse operations in mind.

Localised records with a specific relationship ? for example, customers in the same area or orders that had been made by the same customer ? could be accessed and manipulated quickly because they were located in the same part of the hierarchical tree. If an application needed to conduct operations across a wide variety of unrelated records, however, it would require large jumps across the tree structure, which would slow the database down. This made hierarchical structures inappropriate for operations such as data mining, where the object is to find relationships between disparate records.

When relational databases took hold in the mid 1980s, pointer and table-based infrastructures made the retrieval and manipulation of disparate records much easier. This represented a major paradigm shift in the database market. It has been over 10 years since the commercial acceptance of that paradigm shift, however, and the database industry is set for a reshuffle.

The major feature on the database?s commercial horizon is the object-oriented or object-relational model. This relatively new model is designed to expand accessible data beyond structured, alphanumeric formats into new data types that reflect business needs more accurately. Such data types may include spatial, temporal or textual data types, and companies such as Informix are also offering audio-visual data types, too.

?There are a lot more assets within businesses than you might initially perceive,? says Informix marketing manager Terry Lawlor. ?Those assets take the form of content across the business. Businesses have been making very poor use of all the information that they have, and yet it could be used to help make it more profitable.?

Typical applications that could make use of multiple data types include fraud tracking, where temporal data is linked with spatial data and conventional table-based information to monitor the activities of mobile telephone or credit card users. An application could be developed that is able to track the path of credit card-based transactions and cross-reference it against the value of each transaction, and perhaps even the type of purchase made. The software may then become aware that the pattern of credit card use differs sufficiently from the owner?s previous usage profile to raise the company?s suspicions. This could lead the database to flag the account for the fraud department.

Other potential applications could exploit the ability of object-oriented databases to search for images based on their content, so that a news editor could ask for aviation-based pictures, or graphics with a lot of blue in them. Alternatively, applications linked to dedicated textual data capabilities within an object or object-relational database could be used to analyse passages of text and return all recent news items that presented a negative view of European Monetary Union, for example. Traditional textual search engines can do little more than look for keywords using fuzzy logic, but object-based systems can be programmed with complex business rules to make what are apparently value judgements.

The question is how quickly will this technology catch on. Informix?s problems following its digestion of Illustra?s object database technology have been well-documented. The company reached the point late last year where it decided to separate the dedicated data type capability from the database proper and offer it as a Universal Data Option, implicitly conceding defeat in its attempt to convert the market.

Lawlor protests, pointing out that the object capabilities of its database are being used by roughly a thousand customers. Nevertheless, this figure seems low when one considers the overall size of the database market. The company?s failure to hit home with object technology is compounded by the fact that the technology only accounts for a single-figure percentage of Informix?s overall revenues.

Other database vendors have been more cautious about the object market but have nevertheless posited their own solutions. Oracle released version 8.0 of its database with limited object functionality, and promises more functionality with version 8.1, due out by the end of the year with a Java Virtual Machine (JVM) built into the engine. Nick Gregory, server technologies marketing manager at Oracle, says that although it is possible to extend data types within the engine today, it is not possible to optimise them. That is an 8.1 feature, he says.

While Gregory sees multimedia-optimised databases as a ?blue-sky? feature that will be taken up by the market later on, he is more interested in who will be managing databases in the future. The skills shortage in the software industry will lead to a general outsourcing of applications, he says, adding that the database will go the same way.

?You will probably get big service providers such as EDS hosting most people?s applications,? he says, adding that another driver will be the rise of Internet-based and packaged applications. ?The service provider will then have a high-speed network full of big databases where it will manage your data,? he forecasts.

Sybase?s database product marketing manager Richard Harvey also thinks that object-relational databases will take a while to take off. Like Oracle, Sybase has adopted a softly-softly approach to this technology, relying on third- party vendors to build applications that map new data types on to the Sybase engine, which remains strictly relational. He believes that as databases become more of a commodity in the future, they will need to be more transparent.

Self-tuning facilities already exist in some databases, but these will have to be enhanced and made more generally available, he says, if administrators are to treat the database more as a means to an end, rather than as an end in itself.

Rob Hailstone, ex-Computer Associates database guru and now an analyst at Bloor Research, says that the utopian view of the universal database, containing all of an organisation?s intellectual property and able to answer any type of query, simply hasn?t happened because platforms have become too fragmented. Requirements change too often for companies to realistically construct one big, clean container for data, he says, adding that if this scenario were possible, then the data warehousing market wouldn?t exist. Consequently, the concept of the Universal Data- base, (something which has been adopted by both IBM and Informix in their marketing strategies), is a misnomer, he says.

Instead, Bloor says that one future database trend will be in-memory databases, where the whole of the database is held in Ram at once, rather than being referenced from disk. The advantage of this is that it will dramatically increase storage and retrieval times, he says. However, the price of memory and the niche profile of this market means that it will not be a mainstream activity for years yet, he warns.

?The in-memory concept is interesting because the disk becomes a backup mechanism. If you then organise it using clusters and replication, you may never have a disk copy there at all,? he says. ?The other trend that we?re seeing, which is going to hit the mainstream more quickly, is the concept of putting extra features into the operating system in-stead of the database. It?s the line that Micro-soft is taking with NT and IBM is taking with the AS/400.?

As the market consolidates and operating system owners achieve large market shares with their own databases, this activity will become more prevalent, says Bloor. Adding features to the operating system enables them to be integrated more efficiently, he says, and also means that it doesn?t have to be integrated into the database.

?What you?re starting to see now is a type of middleware that enables you to talk to the database for some operations and the operating system for the other, transparently to the developer and the end-user,? he says.

With its reputation for building features into the operating system, this would be a logical step for Microsoft, although data warehouse technologies product manager Richard Hamblen is much more interested in talking about the potential for low-cost Olap and clustering. The company?s decision to ship its low-cost Olap product (codenamed Plato) on the same CD as the rewritten SQL Server 7.0 database (codenamed Sphinx), is likely to cut prices in the Olap market.

Meanwhile, Hamblen says that as the next generation of clustering technology emerges, offering load balancing capabilities rather than mere mirrored hardware resilience, the potential for balancing database transactions across multiple mach-ines is huge. Don?t be fooled, though ? Microsoft?s promotion of its own transaction server is no coincidence. The company is bound to tie in Microsoft Transaction Server as just another piece of the clustered database pie, and the likelihood is that it will try to accelerate customer buy-in by bundling MTS in with NT 5.0. We?ve seen the future of the database ? and in Microsoft?s case, it?s got homogeneity written all over it.

The power behind Hoodini

Geoff Doubleday, managing director of administration at Nomura Bank, is one visionary who believes in the power of the object-relational database. He has used the technology as one component of his Hoodini (Highly Object Oriented Database In Nomura International) project. Like any good IT professional, Doubleday is simultaneously enthusiastic and cautious about introducing any new technology. ?Our view of object databases is the one that we took at the start of the project in 1995, which is that we couldn?t necessarily depend on a object-oriented database at that time,? he says. ?We couldn?t depend on object data for persistence, reliability and performance. On the other hand we wanted an OO environment.? Consequently, he keeps his static data in a Sybase relational database and then feeds data through on a real-time basis to an ObjectStore object database. ?It?s almost like we have two universes. We have a small one that uses Sybase and then we map it into another, object world and all our systems run off of the ObjectStore view of life,? he says.