How to handle multimedia?
- By Jerry Lazar
- Jul 28, 1996
Thanks to booming interest in the Internet and aggressive marketing by PC vendors, "multimedia" has become a hot button in the federal market. It is seen as a significant factor in the future of training, simulation and document management. How to handle the data products of multimedia is a problem, however, and it's one the relational database world is trying to solve.
There is already some thought of using a relational database management system (RDBMS) to store video on-line for use in virtual reality battlefield simulations, for example.
There are also plans to use a database of audio clips to enhance training capabilities and even to store images in a relational database environment until they are sought under the Freedom of Information Act.
Actually, multimedia - generally defined as a mix of audio, animation, still images and full-motion video - is only a part of the problem. There are many other kinds of data that also will not fit into an RDBMS' rows and columns, such as spatial and temporal data and other non-standard types. And that has forced database vendors to use object technology to create unique solutions.
But there is still a long way to go. It ultimately will depend on how good database companies are at manufacturing a solution. Only recently have they announced products that can handle nonstandard data types, and to do so they have had to rely on painstakingly developed object-relational hybrids, which so far have not convinced everybody of their worth.
At least some observers think multimedia may be suffering from oversell, particularly in the database arena. "Multimedia is an adjunct," insisted John Campbell, president of TSE and Associates, Annapolis, Md., a Defense Department prime contractor. He considers the emphasis on other media types to be less than encouraging.
"I don't want to contribute to the dumbing-down of America," he said. "We are a document-based culture, and I would rather see OpenDoc get moving. The immediate question should be: What is multimedia going to buy me?"
OpenDoc, a developing industry standard, enables the creation of cooperative component software that supports compound documents, can be customized, can be used collaboratively and is available across multiple platforms.
It is the lifeblood of an RDBMS, the rows and columns, that makes these traditional database systems such a poor choice for multimedia and non-standard data applications.
"Trying to define a geophysical space using SQL will drive you berserk," said Peter Kastner, group vice president/analyst for The Aberdeen Group, Boston. "Structured Query Language is just not set up to answer questions like, 'How many cities are north of New York?' "
But that is exactly the kind of question that an object-oriented system can handle, and fairly easily.
Several years ago RDBMS vendors began investigating the possibility of using an object database management system for these kind of nonstandard problems.
It was quickly seen, however, that this wouldn't provide the whole answer.
"Look at a bank: It would be very nice if a bank could model its customers. That's a very object-oriented problem," said Walker White, principal technologist for Oracle Government, Bethesda, Md. "At the end of the day, though, I still want to know how many transactions took place. And that's a relational question, one that is just not very well-suited to an object database."
The next step was obvious: If neither relational nor object would do the trick alone, perhaps a combination would work.
"Relational deals with parts, object with the whole," said Jim August, a product manager with IBM Corp. "Ultimately, you need to deal with both. I can have an object revolution, but I need an evolution of data. I want to marry the image with the card catalog data of that image."
Some database companies, such as Cincom Systems Corp., came to that conclusion more than a year ago, but most of the major RDBMS companies only recently have come out with their products. Sybase Inc.'s Adaptive Server and Oracle Corp.'s Universal Server are just two, and they are designed to provide the speed and efficiency of the traditional RDBMS with the nonstandard data handling capabilities of an object database.
"It's our vision to have a product that doesn't discriminate among the types of data," said Jon Avrach, systems engineer for Oracle Government's defense operations. "We don't care what the data is."
Of course, any system that has the strengths of both object and relational also may have the weaknesses of both. Predrag Dizdarevic, senior vice president, research and development for Computer Associates International Inc., said the company explored a hybrid and then abandoned it.
"We started an extended relational model with Ingres," he said. "But in building the prototype we found we had to compromise on everything. We compromised the relational and lost some of the object capabilities. So instead...we opted for a fully object-oriented environment [Jasmine] and will provide a bridge to our relational system."
TSE's Campbell suggested using caution before going to any relational vendor for an object or object relational database.
"If I were buying an electric car," he said, "the last place I would go would be General Motors...they have too much vested interest in gasoline products.... [For object databases] the answer is a CORBA-2 compliant version of OpenDoc." (Common Object Request Broker Architecture specifies a system that provides interoperability among objects in a heterogeneous, distributed environment and in a way transparent to the programmer. In April, the Object Management Group - an industry consortium that developed CORBA - adopted OpenDoc as its Document Component Facility, enabling the creation of component software that supports shareable and customizable compound documents.)
Campbell has good reason for skepticism. Historically, software companies have a difficult time with follow-ups to successful products, especially when it means a paradigm shift. The switch from hierarchical to relational database management systems in the 1980s was a bloodbath; only a handful of companies survived the transition.
It seems unlikely, however, that many other companies will be joining CA, which seems to have the object-only market to itself, at least among the major vendors. And experience does not suggest too bright a future. Other object-only vendors such as Verdant "have not exactly set the market on fire in the 10 years" they've had the market to themselves, according to Aberdeen's Kastner.
So the hybrid object-relational database management systems have become a Hobson's choice, and the jockeying for position has begun.
Informix, for example, having acquired the rights to Illustra, a software system developed at the University of California, Berkeley, under a contract awarded by the then-Advanced Research Projects Agency contract, has made its system accessible to third parties and other developers in the hope of quickly advancing the case for its solution.
The key to the Informix paradigm is "the ability of the users and partners to plug in their own solutions," said Jackie McAlexander, manager of spatial and scientific solutions and technologies for Informix Federal, Vienna, Va. "We were very careful to design and document in a way for other people to document and support their own extensions."
Kastner applauds that decision, noting that waiting for industry standards - or for one vendor to provide every extension - could mean waiting "the technology equivalent of forever."
Other vendors, such as IBM, have developed multimedia extensions themselves or have licensed or acquired software and put their own labels on the products.
And not everybody subscribes to the notion of systems extensions. Oracle has thrown its weight behind the Universal Server concept, which calls for handling all types of data in a single, integrated environment. On the other hand, Sybase sided with IBM, providing multimedia "snap-ins" for its Adaptive Server.
According to a document from Dennis McEvoy, president of the enterprise business group for Sybase, "full integration is just not one of our top priorities." Rather, he wrote in a recent paper, the company will add data types as the customer requires.
"Some data types - such as time series data - will benefit from the same optimizations we provide for relational data," McEvoy believes. "Others perform much better using specially optimized queries or access methods."
Despite the arguments about what approach to take to the problem, there is no doubt about the potential demand for object-relational database management systems.
"We have a world in which the quantity of electronic data has exploded," said Don Arnold, director of government business for Cincom Systems Inc. "We're not trying to do the artificial intelligence thing: 'Take data and make it information.' We're just trying to take data and make it useful."
"You can expect it to be very strong in command/control/communications systems," Aberdeen's Kastner said. "There are some interesting uses in government-related scientific areas: the National Laboratories, Institutes of Health and NASA. And it would certainly be useful in bringing government information to the public over the Internet."
"You'll already find requirements in a number of areas, going backwards for a few years," CA's Dizdarevic said. "It started with the GIS, anything related to spatial characteristics of the systems. Institutions like the Geological Survey need it because that's the way they represent the entities."
A lot of institutions also are looking to analyze temporal data, he said.
For example, programs such as "the EOSDIS project from NASA, where you are looking for data that has attributes [such as] time, cloud cover and that you can think of in terms of columns, but [for which] you also need access to the images," Informix's McAlexander said.
Text handling and document control may be the most commonly occurring multimedia application today, but it's training and simulation that everyone is talking about.
"If an engineer off-site is putting something together - a really complicated jet engine, say - you might look at a written document," Oracle's Avrach said. "But it would be much easier if you had a video showing what you would find. That could save you hours."
The Air Force also has shown interest in video databases for training - an airstrip can be simulated, and a pilot allowed to practice landing in different weather conditions, or even simulate bombing runs.
Because the video is digitized, the proper clip can be found swiftly, especially if it is indexed carefully. Furthermore, video on a database is available to every user, whereas a tape can be viewed by only one person at a time.
However, it is the intelligence services that seem to have the best idea of how to use these developing database systems.
"They know exactly what they want out of an object-relational database," said Tim Negris, vice president of data management marketing for IBM. "It's classic. As an intelligence operative, you would input textual information and, perhaps, hand-drawn diagrams. And that would, in turn, have to be associated with satellite data and who knows what maps or weather data."
Despite the talk, however, there has yet to be a big push for nonstandard data types within the government sector.
"There is no 'killer ap,' " Cincom's Arnold admitted. "People are still saying, 'I want this because it's neat.' That's because it's still in the growing phase."
Oracle's White agreed: "People are clamoring for it, but they haven't really realized what can be done with it. We have almost crossed the hurdles of delivery before the uses for the capability have been figured out. We have solved the problems associated with getting it to the desktop."
Now that most of the big players have announced their products, this is expected to change quickly.
"The government is starting to realize the value and is starting to pull them along fairly quickly," Kastner said.
To the extent that these object-relational databases are being implemented by the federal government, they are still very much in the pilot stage. "You usually find a significant portion of the population taking the pilot approach," IBM's Negris said. "Historically, that's been out of fear, uncertainty and unfamiliarity. But this is familiar stuff, so the fear factor won't be quite as pronounced."
Still, the last word probably should go to a skeptical Campbell, who warns against jumping in just for the heck of it. "The idea of just opening up and vomiting forth information is not training," he said. "And just take a gander at the price of virtual reality."
Multimedia and other nonstandard data types can be very useful, he said, "but you have to do a cost-benefit analysis. What good is it going to do?"
Lazar is a free-lance writer based in Tenafly, N.J.