InformationWeek

An improperly modeled problem can very easily wield faulty, misleading results. For Dash, the focus is on providing developers with nonprocedural, declarative modeling languages that can render a model in much less code than writing the model in C/C++.

Ilog attempted to ease the pain of modeling in February, when it released OPL Studio 3.0. Among other features, OPL introduces a new high-level modeling language that makes it easier to develop optimization models.

Sabre consultant Sucur used OPL Studio for the first time this past year, while working on a flight-scheduling problem he'd been having trouble with. "Within one week, I'd finished the model I had been working on for six months," he says. Previously, he'd been writing native code that interfaced directly to Ilog's CPlex.

Sucur attributes the productivity enhancements he experienced when he moved the OPL Studio to the introduction of an optimization technique called constraint programming, a technique that comes from the artificial-intelligence world and that has seen widespread use in Europe.

"The really good thing about OPL," says Sucur, "is that it brought traditional models and constraint programming together. This is a big thing."

In the United States, Sucur says, optimization has been dominated by a technique called linear programming, a mathematical approach that solves for every variable in a problem. In the past, heuristic approaches to optimization, such as constraint programming, have been looked down upon by optimization professionals.

Another OPL enhancement is support for the Windows 2000 DNA architecture. With this, Ilog opens the optimization gates to Microsoft Component Object Model developers, Visual Basic developers, and, conceivably, Excel spreadsheet users. This move effectively puts Ilog CPlex algorithms in the hands of any MBA who has taken a linear programming class.

Some longtime optimization users, however, are a bit wary of "easy-to-use" optimization tools. "If you're giving optimization to a wider audience, you are going to get engineers using it who've never worked in operation research before," says UPS's Ware. "That's a double-edged sword. You might easily wind up steering yourself in the wrong direction."

Inexperienced developers frequently misapply their knowledge of optimization and can waste a great deal of time, Ware says. For example, Ware once worked on a problem that would help UPS transition its fleet of airplanes from the slow pre-Thanksgiving season to the busy holiday season. "A seemingly simple problem," Ware thought. On further examination, however, his team determined that to solve the problem, they'd need to run the model as many times as there are grains of sand on the largest beach in California.

The power and complexity--and pitfalls--of optimization are things that business users seem to instinctively understand. If anything, optimization has been limited by people's comfort level with the technology, says AMR's Lapide. "Users have to feel comfortable."

To date, he says, optimization modules embedded in ERP and E-business solutions have not gained the total trust of their users; users often play with optimization features included in products, but stop short of running their businesses off the results.

That's beginning to change, though, Lapide says. "People are getting more comfortable that the computer is giving them better answers."

Ultimately, optimization will be integrated into far more visible aspects of our lives. We are nearing a time, analysts say, when vanilla users will interact with optimization engines via Web-based front ends. For example, optimization might be used for applications such as optimized configuration.

With optimized configuration, the customer might buy a personalized computer online. Behind the scenes, an optimization engine might suggest a model designed to meet a user-specified criterion: price, performance, networkability, etc. Or the optimization engine could help the E-commerce software provide realistic available-to-ship dates.

"We're automating a lot of our decision making," Lapide says. "It's important that the computers we use have some intelligence." And now that we've got the speed, optimization is stepping in to bring us the smarts.

return to page 1, 2