It's an object world out there - Guru Stonebraker leads revolution

Object-relational databases have been fabulously successful in markets where users are willing to try new technology, says database guru Michael Stonebraker, chief technology officer at Informix. 'I fully expect the object-relational market to exceed the relational market within five years,' he says. The early adopters of object-relational include the entertainment industry, media asset managers and digital publishers.

Object-relational databases have been fabulously successful in markets where users are willing to try new technology, says database guru Michael Stonebraker, chief technology officer at Informix.

"I fully expect the object-relational market to exceed the relational market within five years," he says. The early adopters of object-relational include the entertainment industry, media asset managers and digital publishers.

Stonebraker, emeritus professor at the University of California, founder of relational database firm Ingres (in 1980) and the inventor of Informix's Illustra object database, was in New Zealand last week to speak at Informix seminars.

"The big markets in the short term will be GIS, media asset management, law enforcement, government applications, financial services applications for people who want to show time series, drug discovery, and next-generation healthcare systems," he says.

"These are all rich content markets.

"Those markets which don't have rich content now will over time — for example, insurance companies."

Object-relational is also a good option where there are just numbers and character strings, Stonebraker says.

"New Zealand has Maori and English names mixed in a semantically deep way. SQL 92 names have hard-wired ASCII semantics, which are probably not appropriate for New Zealand names. My suspicion is that you need to construct a New Zealand name data type in the engines.

"Very few countries sort names in ASCII order. They use the Julian calendar.

"A lot of ordinary business situations don't obey SQL, so you need objects to deal with them correctly."

The latest product to come out of what Stonebraker describes as "the skunk works" — his team of around 20 developers who have a free hand to develop ideas — is a tool known as Visionary. It's a toolkit for Universal Server applications but not a 4GL.

"Visionary doesn't fall into any of the standard categories of analysis and reporting systems," Stonebraker says. "It covers the full spectrum of data access, dissemination, management, analysis and prediction.

"With Visionary, you can access and format your data without writing code. It has native support for rich data types containing your business logic and other rich data, and it is easier to use than Visual Basic.

"Think of it as a dynamic Powerpoint presentation on the front end of your database.

"Bascially, you get an arbitrary canvas under which you can render any object. You have holes - wormholes - inside the canvas which take you to another canvas.

"There is no limit on drilling down. The user navigates through the wormholes to find information.

"All this comes on the fly out of the Universal Database."

Visionary comprises 500,000 lines of C++ code.

Stonebraker says it's basically very good at decision support where the data needs to be visualised.

"It's in beta now [Auckland University is a beta site] and goes GA at the end of November."

Informix is also reviewing its middleware options.

"There's a fair number of database companies building middleware frameworks but these are simply transaction processing monitors and application servers.

"Our view is that there are enough application servers on the market. Instead, we will play, and play well with the dominant products.

"The big debate is over thick and thin clients.

"The thin client runs thick middleware. However, the thick argument totally misses the point because object-relational allows you to run in the system.

"It all depends on the database. Where you want to run the components depends on what you're doing"

Stonebraker says what is needed is universal components — write once, deploy at any tier, and be able to move them if necessary. "The component argument has to be extended to the database."

He predicts there will be universal protocols over the next couple of years.

Join the newsletter!

Error: Please check your email address.
Show Comments
[]