June 1996 € Volume 23 € Number 3

Planning for the Future

Network optimization positions NYNEX for the 21st Century

By David Greenfield

This is another in a series of articles based on interviews with recent Franz Edelman Award finalists. Geared toward practitioners, the articles strive to provide lessons the Edelman authors learned and some pitfalls they encountered during the course of their work.

The Edelman competition, sponsored by the College on the Practice of Management Science (CPMS), the Practice Section of INFORMS, annually recognizes and rewards outstanding achievement in the practice of operations research and the management sciences. Persons interested in more information on this or other recent Edelman work are encouraged to read the annual special issue (January/February) of Interfaces in which the full papers submitted in conjunction with the Edelman competition are published. To obtain a copy of the most recent special issue of Interfaces or for subscription information to the journal, call 1-800-4INFORMs.

Connectivity is the operative word for most everyone in the information age. No company can afford to be an island as firms of all sizes get wired for business both locally and worldwide. And with the growth of the Internet connecting us all faster and more graphically than many would have imagined just a few years ago, the immediate and long-term future of business operations will be dictated via electronic commerce of one sort or another.

At the heart of this connectivity is telecommunications. Without the advancement of phone line networks, through which the lifeblood of the worldwide networks flow, the burgeoning growth of our increasingly connected world will be severely stunted.

A prime example of the advancing technologies being implemented by the telephone companies to keep up with the increasing demands put on their systems can be found at NYNEX, the regional Bell operating company for the northeastern United States. Seven years ago, NYNEX began looking for a new way to improve planning for its interoffice facilities (IOF) network -- the backbone of the company's system.

The IOF network is comprised of the fiber optic links which connect NYNEX's 1,200 sophisticated digital switches. Investment in this part of the network is crucial, as efficient IOF planning ensures NYNEX's future position in the market and helps the company achieve the maximum benefit per dollar invested. To facilitate IOF network planning and keep it running smoothly, NYNEX sought to automate the planning process by creating an expert system that would minimize cost and satisfy all planning constraints while applying corporate policy. The system developed to do this is known as Arachne.

The Arachne project was driven by two major factors:
1.) Expense-driven costs. The old method of planning was very man-power and time intensive. By the time one network plan was completed, it was time to start another one. So people were only able to deal with the "hot spots" -- places likely to produce problems in providing capacity in the network. There was no time to look across the network at emerging issues or to deal effectively with the integration of new technologies into the network.

Rangu Salgame, executive director of Network Planning and Engineering Systems Laboratory at NYNEX Science and Technology, says that the old, manual method of planning for the IOF network took six months for one planning cycle. "So with two planning cycles a year, the planners didn't have any time to go back and look at the plans in order to do a more efficient optimization of the plans they generated. And because they typically had to do the planning one piece at a time, they were never able to do a global plan. Considering the size of the network we have, it was just too complex to do manually," he says.

2.) Capital effectiveness. Jim Euchner, vice president of Process Improvement R&D at NYNEX Science and Technology, says, "NYNEX spends hundreds of million of dollars a year on interoffice network facilities, and the idea behind the Arachne project was that if you could improve that expenditure by even a small percentage, it would be well worth it. When we went back and compared the manual planning method versus the use of Arachne, the itemized capital savings from Arachne came out to be more than $30 million. The savings came from identifying network elements that could be disconnected and deployed elsewhere because they were underused, and from optimizing the embedded base to reuse existing facilities as opposed to installing new network elements. It has also allowed for the rapid implementation of new technologies by allowing us to plan for the use of such technologies across the entire network."

Teamwork
The internal parties behind Arachne's development were NYNEX's network planning groups, which worked as a team serving as domain experts, and NYNEX's Science and Technology labs, which performed the knowledge engineering and software development related to Arachne. No outside experts were used at any point.

Facing these teams was the initial business problem that faced NYNEX IOF planning itself: How do we better use capital to more effectively bring in new network technologies and augment the network to meet the changing marketplace needs?

Salgame was quick to point out that NYNEX management was behind the teams' ideas from the start, but constantly had to be reminded of the feasibility and effectiveness of the system. "A very important aspect of developing a new planning system is that you have to understand the business issues that the company is facing and, rather than selling the technology or the solution, be able to convince the management of the effectiveness of the solution and measure the results. That was the approach we took, and it was a critical success factor for us," he says.

With management behind them, the Arachne teams used a two-key technology to develop their new, automated planning system. The first of these keys is expert systems technology (artificial intelligence); the other is optimization techniques. The expert systems portion assured compliance with planning rules-of-thumb and NYNEX policies; the optimization algorithms found the least cost routes and increased utilization of the network. Both the knowledge base and the algorithms were developed internally.

"Certain pieces of the system ran on expert system technology and other pieces ran on a hybrid integration of expert systems and optimization techniques," Salgame says. "And that was the uniqueness of the system -- bringing together two different technologies to solve one business problem with different facets of the problems solved by particular technologies."

Ensuring that the teams worked well together was an important point in getting Arachne developed on time and in a workable fashion. The person behind the success of the teams and who had the initial business vision for the project to begin with is Bruce Spinney, who at the time was the head of the Network Planning Group at New England Telephone, a division of NYNEX.

Spinney, currently the director of Engineering and Construction for Rhode Island at NYNEX, says that his vision for the system came from looking at the corporate issues the firm was dealing with at the time. "We were not able, with all the mechanized help we had, to get the planning job done on time. It was very labor intensive and capital intensive to get our product out and it was always late. We looked at all our mechanized data, but there was no way to pull it all together. Most of the data in the system involved repetition. So we thought about what kind of tool we would need to go in there and develop some kind of generic rules in order to take care of the repetition. And that's really where my dialogue with Jim Euchner began. I had the business problem, and knew the data and what we needed, and I also knew the business rules we would need to make it work; what I didn't have was the mechanized background or the expertise to really pull it off. So that's where we started."

At this point, Euchner brought in his Science and Technology experts using rule-based processes to develop a prototype which told them the project Spinney had in mind was feasible.

"A lot of the teams' management was handled between Jim and I," Spinney says. "I always had my best people on the project -- those with the most knowledge and the ability to look ahead -- and Jim had his best people there as well. But you had to break down the wall. We spoke different languages. There was a major communications problem about what we needed to have happen and what the computer scientists wanted to happen. So Jim and I just had meeting after meeting so that eventually each side could talk with each other. If there were unresolved issues, no decisions would be made until the teams sat down and worked through it together."

The communication walls Spinney refers to were not of a competitive nature, rather they were representative of two different communications schemes. "It was difficult for the end user who uses a keyboard and understands the network but doesn't understand anything about the computer science that's behind it," Spinney says. "And the computer scientists didn't understand the network as we did. We had to educate my people (facility planners, who were not computer literate) so they would be aware of what the computer scientists were doing and vice versa."

To make sure these meetings worked at breaking down the barriers between the two groups, Euchner and Spinney never missed a meeting.

"People who reported to me knew this was an important part of the process and that we had to make it happen, and the same went for Jim's people. Jim and I took a very parochial approach to the getting the teams working together more efficiently," Spinney says.

A beneficial step that Euchner took in the development process between the teams, according to Spinney, was to include Spinney's people in all the testing from the prototype onward. Spinney's teams were able to watch Euchner's team as they wrote the code, and were allowed to review the data at each step.

Expert Resources
Among the teams, a tremendous amount of knowledge regarding the IOF network abounded. "A lot knowledge existed about the infrastructure, the planning process, the idiosyncrasies of the network and specialized knowledge about different pieces of the network and how best to do the planning" Salgame says.

And it was in regard to this specialized knowledge that the expert systems approach proved particularly beneficial to Arachne, Salgame adds, because it "allowed us to leverage the knowledge of the best experts in the company in order to use it for better decision making. At the same time, however, expert system technology doesn't lend itself readily to optimization. These expert planners were able to perform local planning very effectively, but were not able to do global optimization. Traditional OR optimization techniques are very suitable to solving such global optimization, but are weak in capturing the intricacies and knowledge in the problem. So by bringing both technologies together you get the best of both."

But the implementation of expert systems required some initial groundwork. "Early on in the process, when we were doing the knowledge engineering (KE), there were some surprises," Euchner says. "For example, we used KE to determine the rules for developing the system to plan the network, and we agreed on these rules. However, when the rules were actually applied across the entire embedded base, the network planners were aghast at the result. So the rules had to be modified to make the overall plan reasonable from a cost perspective."

The expert system technology used in Arachne was developed in-house, but an expert systems shell was used (KEE from IntelliCorp.). Euchner added that NYNEX has since migrated from the use of KEE to C++.

From a knowledge engineering perspective, the Arachne system utilizes very simple rules at the lowest levels, although they are exhaustively applied. As you move up in the hierarchy, however, the rules get more complex and are integrated with optimization algorithms.

After developing the prototype for the system, Euchner wanted to take it back to NYNEX and develop a commercial product for Spinney to use, but Spinney contended that the prototype brought his division so much benefit that he didn't see any reason to wait. This variance of opinions laid the groundwork for the biggest point of contention that arose between the two teams during Arachne's development.

"Our point," Euchner says, "was that this, at the time, was only a prototype and it would require a lot of support and hand holding. And we thought we should really go back and commercialize the code first."

"Any time you do a prototype you have support concerns," Spinney counters. "But the 'handshake' between Jim and I over this issue made it clear that if it was my decision to use the prototype, then I could not go back and complain to him. I had to take the good with the bad, and we lived up to our end of the deal. We had some frustration with prototype problems, but the simple fact of knowing that what we were doing now with the prototype was so much better than the way we had done it before made it acceptable. And because the teams had worked so closely together during the development, a lot of the tweaking required for the prototype was able to be done over the phone because we knew each other pretty well by then."

Despite all that the teams had been through to bring Arachne to the point of commercialization and use, the biggest challenge remained -- implementation. Implementing Arachne at NYNEX involved working the system into the organizational and business processes of the corporation, which meant integration with the legacy systems and the corporate databases.

"By doing this," Salgame says, "we have created an integrated system and not a stand-alone silo. And we addressed this concern at the very beginning, in fact, by automating the process of data retrieval from the mainframe corporate databases, bringing it into the Arachne system and, once the planning was done, pumping it back into the corporate database automatically."

In sum, the whole Arachne planning system is about integration and evolution. "First came the integration of technologies, second was systems integration, and third was the ability to evolve the system," Salgame says. "These steps were important to have built in to the system because knowledge is always evolving within the corporation, and the challenge is to capture that knowledge within the system, thereby allowing it to progress with the business. Through the development Arachne evolved a process by which we constantly worked with the network planners to keep the system growing and able to accept changing technologies."

Future Use
The ability of Arachne to open up the IOF network to changing technologies bodes well for the applicability of the network. "We have not yet leveraged the system as far as possible," Euchner says, "because Arachne can plan in two weeks instead of the six months that we have long been accustomed to.

"We also haven't yet extended Arachne to permit the testing of alternate scenarios for doing strategic planning. This hasn't been done because the system needs to be architected to enable more easy what-if games and operate more quickly. We've been working on that. But we have been satisfied with the system as it currently stands and we know that it can be taken much farther."

The system, as it is now designed, is pretty open in terms of what changes can be made, Salgame says. "We have put a knowledge change process in place where we work with the end users and have two releases each year through which we plan for changes six months in advance. Then we put a team together from planning and the Science and Technology labs that work together to implement the changes that need to be brought into the system," he says.

Euchner says that NYNEX is now working on speeding up the algorithms so instead of it taking several hours to plan the highest network level, it can be planned in minutes. "We are also currently attempting to determine a level of network goodness," he says, "so that we can move on to a second level of optimization."

David Greenfield is the managing editor of OR/MS Today.

For more information on this article, input the number 1 in the appropriate space on the Reader Service Form

Click here to return to the table of contents.