Newsmaker: IBM's advance team does the math

(continued from previous page)

Is your job to act as a sort of a SWAT team to handle really hard problems?
Pulleyblank: In fact, when we set up the mission of this center, part one of the mission was to solve really hard problems. It's the "Star Trek" part--go boldly where no one's gone before. That's cool and that's what the research people love.

But the other part is absolutely critical. I spend more time worrying about this than any other part, and that's to make it real. Find a way to take that particular capability and deliver it to clients so that we can get into their business operations effectively. That's been the biggest obstacle to the use of these kinds of methods in businesses until now--they have been too hard to use, too inflexible. So for a company to optimize the marketing campaign, first thing that you should do is hire 25 Ph.D.s and leave them alone for three years, after which they may know what to do.

If that's what they do, they'll never get to it. So, we're trying to short-circuit that and find ways to speed it up. Then the third part is simply to turn it into a sustainable, profitable business for IBM.

Part one of the mission was to solve really hard problems. It's the "Star Trek" part--go boldly where no one's gone before.

How does your background as a researcher in high-end computing relate to this?
Pulleyblank: There is a company called SmartOps, that is a small company that does inventory optimization, which came from Carnegie Mellon University. They've done some inventory optimization with some major companies. One particular problem had 70,000 SKUs (stock keeping units). Doing the inventory models on it took six hours for them to run it. Because they have a university background, I think they were wondering what would happen if we put it onto a Blue Gene supercomputer. So we enabled them to do that, and they just used one rack of Blue Gene and it went down from the six hours to 17 seconds, which solved the problem.

If it takes you six hours to solve this problem, you do it overnight and you come in the next day, and hope you got an answer. If it's 17, 18 seconds, you do it, you look at the answer, you change something, you do it again. So all of a sudden these become operational, and that's the other enormous change. It used to be that this math stuff was used for planning, which meant you couldn't use it in operations because you couldn't do it fast enough. But now we're saying we can do it fast enough that it'd be part of the operations.

IBM is creating development centers around the world, including India, with the goal of creating replicable "solutions." Is that part of your plan?
Pulleyblank: When I'm working with a client directly somewhere in the U.S., for example, it will be mainly people on the ground who are working with the client on all kinds of areas. What happens then is as we start to turn it into a solution, which we're going to replicate and reuse, we'll architect it and get it rolling for the teams in India who'd be the ones who would do a lot of the development for us on that. So in particular, some of the work we want to do to make it effectively consistent with the standards software stack, standard architecture, this will be the kind of functionality which they'll be able to do for us very effectively. Now, we'll see how that evolves, but that's where we're going initially at least.

Isn't this reuse idea kind of the holy grail of the IT industry--making software production more like a manufacturing processing by reusing components?
Pulleyblank: Think of it. Who is the biggest company in the software industry? Microsoft. How do you become the world's richest person selling something that's only $150 (like Windows XP)? You sell a lot of them. You see the software industry has been largely built around high volume, lower unit price.

When we go into it, we're going the other direction. If we look at these services-led asset-based solutions, we're not going to sell millions of things, and they're going to be much more expensive than your few hundred dollars, or few thousand dollars, just for the asset itself. It's quite a different model because it's different types of problems that it's going after. So, in some sense, there's a whole part of the software industry that is on a completely different trajectory from we're doing

Doesn't that limit your market, though, to the really hard problems in the world? Presumably, that would mean a smaller set of customers.
Pulleyblank: It may not be everybody, but that's OK. I hope some of the things that we develop will in fact evolve into standard software products at some point, and be very broadly used. And when that happens, it'll be a group like the software group within IBM that will take them up, and that's fine with me. The spreadsheet is an example that has become absolutely mass market. Even though it's a very sophisticated thing, it evolved into that.

That's what we do--that's our particular center. The thing I stress is we do want to get the hard problems, but we have to do it in a way so that people can use it, and it's that dynamic which really drives a lot of what we do.  

More Newsmakers