Waiting for an Internet connection is worse than waiting for a kettle to boil. At least with the kettle you do not while away your time watching flashing on-screen advertisements.

A research team at the Dublin Institute for Advanced Studies has hit on a way to cut down on the wait, however, by controlling congestion on high-speed telecommunications networks. It could revolutionise the way the Internet works, according to the leader of the group, Prof John Lewis.

No one predicted that the new Internet services coming online would generate the traffic explosion they did, Prof Lewis said. Coming in rapidly behind them are more services hoping to crowd onto the same transmission lines, including local area networks, satellite communications, cable television, teleconferencing and two-way games and video services.

All these have been helped onto the market by the tremendous bandwidth, or carrying capacity, offered by fibre optic technology. Even these broadband networks are under pressure because many of the fancier services such as video gobble up bandwidth, and too many people on a service at one time can also make bandwidth disappear.

The great challenge for the "teletraffic engineers" is trying to manage these high-speed networks, Prof Lewis explained. And an inherent problem for them is trying to deal with the demand. There can be Everest-high peaks in demand followed minutes later by deep traffic valleys. Giving too much bandwidth would be wasteful and expensive but if you provide too little then users will have a poor quality service.

Prof Lewis, who is director of the School of Theoretical Physics at the Dublin Institute for Advanced Studies began thinking about some of these problems. It seemed a worthy subject for the School's Applied Probability Group (APG), which involves staff and graduate students who combine fundamental research in mathematical physics with the development of practical solutions for the telecommunications sector.

The APG, he said, "is very informal, we started calling ourselves that in 1990". It currently includes two assistant professors and 10 students. "What I insist on is these graduate students do their degrees in basic research. They spend half their time in research and half their time on the project. We are pursuing things which we think are interesting in probability theory and information theory. This is the area where there is work going on."

The APG has been very successful finding ways to direct its basic research into real world applied situations. It has several contracts with Swedish telecommunications company, Telia, who are also strategic alliance partners of Telecom Eireann. Telia has also set up a separate company here, Measure Technology Ireland Ltd, (MTI) to commercialise research coming out of the APG.

One of the most popular ways to move data of all sorts about on these networks is to break it up into discreet parts of a uniform size called cells. Each 53-byte long cell includes an address where the cell will eventually be delivered. Even if a signal gets jostled or its cells get separated, all the data eventually gets there, sent on its way via the network's switches which read the address and pass the cell down the line.

During the early 1990s teletraffic engineers began studying what was happening on overloaded networks. They were trying to find ways to level off the traffic load, shaving off the hills and filling in the valleys as a way to reduce congestion. They reasoned that while the peaks could cause problems, perhaps they could push some of this demand into the quieter moments on the network.

One way to do this was to produce buffers where the cells which made up a signal could be stacked for a time, waiting for a chance to be processed by the network switch. The network would hold back messages until traffic cleared and then the buffered signal was sent on its way.

But buffering presented two problems. It could produce unacceptable transmission delays and if the buffer became too full, then data would "overflow" and be lost. To reduce these problems the engineers introduced Connection Admission Control, monitoring software that determines whether the network has sufficient capacity to accept new connections and signals.

Most CACs use complex parametric modelling to decide when there is too much traffic but the models are slow and have problems reacting fast enough to deal with burstiness.

"It takes an enormous amount of effort to do these models," Prof Lewis said. You end up with too much information about traffic and not enough about the buffering and the risk of lost data which is the object.

The DIAS group seems to have found a way around this problem. They don't try to model activity in the buffers, they discovered a clever way to read it directly and in real time. "The parametric modelling of traffic was totally unnecessary and wasteful." Their system gets an immediate read on what is happening in the buffers and this information can be channelled back for example to the CAC which can be told in real time whether the road ahead is congested or clear.

Their approach was based firmly in theoretical physics, appropriately enough. They first published their work in 1995, with results that were so promising that they attracted the interest of Telia and the Cambridge University Computing Laboratory. By 1996 the three partners began an EU-funded study and before long patents started to arise.

Telia established MTI last year specifically to commercialise this technology and it already has a customer in telecommunications manufacturer, Ericsson. The technology is also being demonstrated to switch and router manufacturers in the US.

The APG also has a troubleshooter contract with Telia, with its engineers throwing up problems and the APG trying to solve them. "The topics on applied research come to us from Telia engineers." This is just the way that Prof Lewis likes things. The students must do their basic research to get their degrees, but "we let them cut their teeth on applied problems".