By Sheila Nason
Once upon a time,
perhaps a dozen years ago, families communicated by letter and phone,
engineers flew across country to go over designs with clients, and almost
no one had ever heard the term dot.com.
Szymanski and Vastola
are members of an interdisciplinary group of Rensselaer faculty members
who have taken on that challenge. Their job is to see that the rest
of us can access Web sites, download massive files, and e-mail documents
around the world without slowdowns or crashes.
MAKING A DIFFERENCE
These researchers are winning respect by developing technology that is already making an impact. They have been funded for every grant theyve sought from the highly competitive Defense Advanced Projects Research Agency (DARPA), the agency that financed the original development of the Internet and now is pushing scientists to create the technology that will support the Next Generation Internet (NGI). Since 1995, three DARPA grants to Rensselaer have totaled more than $2 million, with Rensselaer proposals regularly winning funding and helping shape the direction of DARPA research. Kalyanaraman, moreover, is an active member of the Internet Engineering Task Force (IETF), a self-organized, independent group that develops standards and protocols for the Internet. His research already has impacted some current standards and now he is involved in creating standards for multicast and congestion control. He is also consulting for both major and start-up companies. Kalyanaramans influence was recognized this year by MITs Technology Review, which named him one of the top 100 Visionaries for the Millennium, those people who have the greatest potential for technological innovation in the 21st century.
The strength of the entire networking group, in fact, is a major building block in the Rensselaer Plan, the Institutes new strategic plan that was approved May 12 by the Board of Trustees. The Plan, designed to move Rensselaer to greater prominence in the 21st century as a top-tier world-class technological research university, calls for substantial investment to create new constellations of excellence in information technology and biotechnology.
In information technology, Rensselaer will build on its strengths in networking and multimedia communications to establish new constellations of world-class faculty, staff, and students in such emerging areas as pervasive computing, tetherless systems, and distributed sensors, explains Arthur Sanderson, vice president for research.
AN INTERDISCIPLINARY TEAM COMES TOGETHER
Vastola joined the Rensselaer faculty in 1983, recruited by Rensselaers world-class communications group, which had realized that networking expertise was increasingly vital to the ECSE Department. Since then, his research has focused on network management, wireless networks, multimedia, and network design and analysis. He has been an editor of two leading networking journals, IEEE/ACM Transactions on Networking and Computer Networks, and he was technical program chair for IEEE Infocom: the Conference on Computer and Communication Networks.
[In 1985, Vastola and Martin Schoffstall 82 co-advised an ECSE masters student who wrote the first Unix implementation of the Simple Network Management Protocol (SNMP), which Schoffstall had co-invented. SNMP is the basis for much of the teams network management research.]
Vastola compares most network users to automobile drivers who want to turn the key and go, without having to study details of how an internal combustion engine or a transmission works. Most computer users want to hit the send key and have their e-mail delivered instantly without worrying about how it gets there. Our work is very much inside the network. We focus on the nuts and bolts of how the Internet works, he says.
Szymanski joined Rensselaers then newly founded computer science department in 1985 from the University of Pennsylvania. His research focused on high-performance computing systems. For his work on parallel and distributed systems, he was elected an IEEE fellow and became editor-in-chief of the international journal Scientific Computing.
In 1995 they teamed up with Wes Kaplow 98 of Lucent who was Szymanskis doctoral student.
That same year, a $1.4 million grant to develop Proactive Network Problem Avoidance was awarded by DARPA to a team that included Rensselaer, Lucent, and Penn State, with Rensselaers share totaling $500,000. At the time, DARPA was putting together its Next Generation Internet (NGI) program, an initiative which was influenced in part by this grant proposal.
In 1997, Vastola and Szymanski recruited Kalyanaraman, an expert in traffic management, to join ECSE. Vastola calls this one of the most outstanding hires in Rensselaers history.
Szymanski had used simulations to solve problems in ecological systems, economics, and computer systems. The team began considering simulations to find cures for network traffic problems. Szymanski likens the process to a doctor who cultures a sore throat and then tries different antibiotics on the cultures to see which work best.
Szymanski proposed to use a decomposition method that he became familiar with while working with Nobel Laureate Lawrence Klein on econometric modeling. When applied to networking, the method calls for breaking the large network (or even the entire Internet) into domains and running simulations of each separately. In the first iteration, interdomain traffic is ignored. In the following iterations, it is approximated by the traffic generated by each domain and directed to others. By carefully monitoring convergence of these iterations, the correct solution for the entire networks can be found.
The method is efficient because the simulation time of a network grows faster than its size, explains Szymanski. If you divide the network into 10 parts, simulation of each runs 100 times faster than the simulation of the entire network, so you have time to iterate with partial solutions until the convergence to the full solution is achieved.
This idea, combined with the Kalyanaramans method of using fast simulation for traffic control, Vastolas ideas about simulation modeling and routing control, and Jis work on traffic modeling, won the networking team its second DARPA grant, bringing more than $600,000 to Rensselaer. Their novel approach to network simulation and its use in network management influenced DARPA to create a new program in Network Modeling and Simulation. The programs goals are close to the research directions of the Rensselaer group. As a result, this summer the team won its largest DARPA grant, $950,000, under the NGI program.
SCALING UP AND SPEEDING UP
The second DARPA grant proved feasibility of network management based on simulation: the simulations identified possible cures, and routing and traffic management techniques were automated to implement the cures. With the third grant, the team is trying to scale up its work to large networks or even the entire Internet. In addition, they must speed up their simulations, since answers that come too late are useless. The astounding goal is an improvement of five orders of magnitude over existing technology.
The team also needs a much faster simulation engine. Christopher Carothers, who joined Rensselaers computer science department in 1998 after earning a doctorate at Georgia Tech, is helping build this engine.
Weve constructed ROSS, Rensselaers Optimistic Simulation System, which has the power to be extremely fast in simulation of simplified sample network models, he says. Now we are looking to see if we can use that and ideas from other people to create something that is better and faster, and also usable.
Carothers is collaborating with researchers at Georgia Tech, UCLA, Dartmouth, and other DARPA-funded groups to let them know about ROSS and to learn about their research.
Even with such improvements, it will not be possible to run all possible simulations. If there are a million possible solutions and you can only afford to simulate 1,000, there must be some way of choosing. Kalyanaraman, an expert in experiment design, is tackling this question. Vastola is using advanced modeling techniques to simplify the model being simulated. Together they are demonstrating that this approach actually solves network problems using traffic management and routing.
Network communications are so central to so many fields that the team is involved in numerous collaborations both on- and off-campus. Collaborations with Rensselaers image processing and video transmission group have developed, for example, because the networking team is working in areas such as video conferencing that rely on network transmissions.
Ji has been on sabbatical this year at Lucent Technologies Bell Labs and at MIT, working on intelligent agent technology that can monitor entire networks instead of single nodes. Earlier, Jis agent was looking at the router at one network node. This is like having a policeman at one corner of the city during rush hour. Now the goal is to build a multi-agent system, operating like hundreds of policemen monitoring widespread corners and communicating with each other to determine how traffic should be rerouted. She also is cooperating with a team at MIT that concentrates on the physical network, including wireless and optical systems.
Kalyanaraman, meanwhile, has research grants from Nortel Networks, Intel, Packeteer, Pulsecom, and Reuters, and is talking to other major companies in his determination to stress immediate technology transfer.
Szymanski is funded under the NSF Knowledge and Distributed Intelligence program to investigate network-based simulations and use of agents to link simulations over the Internet. In addition to Rensselaer, this project involves researchers from Dartmouth. For many years, IBM has funded Szymanskis work on use of agents for network performance collection and network anomaly prediction using neural networks. His students involved in this research won prestigious IBM Fellowships and a Gates Millennium Scholarship.
Vastola is also funded under a Department of Defense Multidisciplinary University Research Initiative (MURI) in wireless communications. In addition to Rensselaer, this wireless group, which is headquartered at Dartmouth, includes Harvard and Illinois. Vastola was on sabbatical earlier this year at Harvard, focusing on this project.
Technologies such as video, wireless communications, and networking are converging, Vastola says. If all of the information packets involved in a video transmission dont arrive on time, the picture freezes or jumps around on the screen, providing an unacceptable quality of service. The future is multimedia anywhere, anytime, and designing networks to deliver this is our challenge, he says.
While lost messages, inaccessible Web sites, and jumpy video are often little more than annoyances today, they will grow into major roadblocks to economic growth if better ways to manage network traffic are not found. At present, a small group of experienced network managers deals with problems as they develop, based on what has worked in the past. Every projection indicates that as network use continues to expand, there will be a tremendous shortfall of such people, Vastola says. It would be far more cost-effective and reliable to automate the process, he says. In addition to numerous doctoral students doing advanced research, an impressive group of masters-level students have been recruited to take the research of the group to the next stage by building prototypes. The value of this experience is made clear by the job offers the students receive, with starting salaries averaging $75,000.
The final stage of the research is bringing this research to production. With strong encouragement from Rensselaer trustee Paul Severino 69, the networking team also is looking at launching an entrepreneurial venture to commercialize its technology.
A REVOLUTION IN THE MAKING
One description of the need for further network research is found in Rensselaers latest DARPA proposal: Without the proposed work, network management and control will have to rely on ad-hoc methods, experience of the network managers, and partial evaluation of a few scenarios by simulation. This will critically limit the reliability and performance of large-scale networks, especially in the face of rapid changes resulting from the deployment of new resources, reassignment of human resources geographically, natural disasters, terrorist attacks, and other military threats.
In addition to network efficiency, there is no question that adding bandwidth, using it efficiently, and improving reliability will revolutionize almost every aspect of our lives.
AMR Research Inc. in Boston estimates that business-to-business e-commerce will rise from its current negligible level to $5.7 trillion by 2004. Jupiter Communications Inc. published a study this summer predicting that U.S. business-to-business online commerce will reach $6 trillion by 2005. These trends drive demand for network management technology.
Robert Graves, professor of decision sciences and engineering systems, directs Rensselaers Electronics Agile Manufacturing Research Institute, which was founded to help electronics firms function in the new economy. New circuit boards now can involve designers in one company, board fabricators and assemblers in others, and parts suppliers scattered around the country, all com-
municating over the network.
Greater bandwidth gives more reliability. With more confidence, new applications will be developed, Graves says. It once was not possible to exchange information in massive design files or databases in a reasonable time, but increased bandwidth is making such communications possible, he says. This should bring drastic reductions in paperwork, improved levels of interactivity between companies, and less people time. For customers, this will mean lower prices, higher quality, and a greater selection of custom designs, Graves predicts.
BIGGER PIPELINES NOT ENOUGH
In response, consumers everywhere are connecting to bigger pipelines. Rensselaer, for example, is now part of Internet-2, reports John Kolb 79, dean of computing and information services, with an OC-3 (optical carrier) line available for researchers. The T3 line used by most of the campus is being replaced this fall by a second OC-3 line. The campus is going from a capacity of 45 million bits a second to 310 million bits a seconda sevenfold increase.
But bigger pipelines alone do not solve all the traffic problems, cautions Kalyanaraman. Even the widest freeways have traffic jams, when demand exceeds capacity at certain points, when accidents block lanes, or when smaller off-ramps cant handle the traffic pouring from the bigger roadway.
On cyberspace roadways, the problems are formidable. Mismatches have escalated, with terabit (trillion-bit) systems sending messages to megabit (thousand-bit) lines. Inefficiencies can slow traffic in the widest lines, and sudden surges of demand can slow or bring down a network. A recent study done by Szymanski and his students showed that the number of routing anomalies in the Internet stayed the same over the last five years, despite improvements in network hardware.
Most of us have no more patience with these slow-downs than we do with a pileup on the freeway. The Rensselaer networking team is learning how to forecast and decrease network pileups and to route traffic around the problems when they occur.
Our goal is to make the Internet run more efficiently, delivering more of your messages and the services you want in a timely manner and at reduced cost, Vastola concludes.