Internet heading for a fall, says founder

Internet heading for a fall, says founder

Arpanet scientist warns of coming network crunch

The internet in its current form is unsupportable and dangerous, according to the man that led its inception.

Dr Larry Roberts led the team that designed and developed the web's precursor, Arpanet, and is now chief executive of router vendor Anagran.

Dr Roberts warned that poor programming means that the internet is growing too fast for network infrastructure to cope. In addition, security is weaker than it should be because of a lack of development within the TCP/IP standard.

"It has been 40 years and nothing has changed," he said. "Traffic is growing faster than network costs are dropping. Moore's Law is not keeping up. The network costs double every three years, and we cannot afford it any more."

Dr Roberts maintained that the industry needs to concentrate on sending prioritised flows of data traffic rather than sending packets individually.

Doing this could cut the power consumption and size of network hardware by a factor of five, according to Dr Roberts, which would buy some time.

He added that there are big problems with security in that user authentication needs to be checked automatically, and the source of packets needs to be checked.

Data flow records also need to be kept, which would help in dealing with denial of service attacks.

Dr Roberts believes that it would be relatively easy to make the internet secure, but that no-one is willing to do so because of the expense.

He is also in conversation with governments over plans to allow the emergency services to get priority on internet traffic.

Fairness is also an issue, according to Dr Roberts. Peer-to-peer users take up to 80 per cent of the available bandwidth at times, which is neither fair nor rational.

But others disagree. Professor Lutz Heuser, chief development architect at SAP, said: "We have stretched TCP/IP as far as we can go. Do we really want to use 40 year-old technology?"

But Roberts insisted that TCP/IP is sound, and just needs work on its implementation.

"The problem is that we have ignored research into networking because we thought that the problems were solved. They are not," he said.