Threats to the Internet: Too much or too little growth? Andrew Odlyzko University of Minnesota http://www.dtc.umn.edu/~odlyzko Is the Internet about to collapse from a flood of traffic? Or is it suffering from too little traffic? The growing media hype, with its frequent references to exafloods of video choking the Internet, uniformly points to the former. But Internet traffic growth, while still vigorous, has been slowing down, and the latter may be the more serious threat. The rate of growth on the Internet is critical for the future of the core network, for the backbone providers and their suppliers of fiber, optoelectronic systems, routers, etc. (It is not as critical for other segments, since some of the most profitable services, such as SMS, and wireless in general, generate very little traffic.) Which technologies will be needed depends critically on this growth rate. As just one example, Larry Roberts, one of the "Fathers of the Internet," continues his long and so far futile campaign to prove that his creation was a mistake, and that we should move back towards circuit-switching approaches, . There are arguments that his solution may not be the right one, . But even aside from that, it is questionable whether the problem that Roberts identified exists. His claim is that Internet traffic is growing at about 100% per year, and that router technology cannot keep up. But the evidence we have, although not definitive, strongly suggests that traffic is growing at only 50-60% per year both in the U.S. and worldwide, . If those rates continue, the problem that Larry Roberts poses may not materialize. The danger in settling for low growth rate projections is that we may not have enough transmission capacity to handle new demands. And, as it outlined at , there are huge potential sources of additional Internet traffic. As just one example, there is already far more stored data as well as broadcast video than is seen on the Internet. And the past should have taught us to anticipate the unexpected, such as the browser, P2P, YouTube. Thus there is certainly an argument for having ambitious plans. Even today, without any new technologies or services, we find traffic levels in places like Hong Kong and South Korea that are about 6 times those of the U.S. on a per-capita basis. The danger in accepting exaggerated growth projections is that we may end up with another painful glut. However, the glut is not likely to be as severe as the one early this decade that led to the telecom "nuclear winter" earlier in this decade. There is no sign any more of those preposterous "Internet doubling every 100 days" stories that John Sidgmore and Mike O'Dell used to delude the industry with. The highest rates one ever hears projected are occasional mentions by John Chambers of Cisco of 5-6x annual growth rates in the future, and that seems to represent just wishful thinking. The 4x growth cited in a recent report is not credible, since it relies on a projection made by Larry Roberts in 1999, and results in current traffic levels in the U.S. of several tens of megabits per user, clearly nonsense. Most reports project far more modest growth rates. Larry Roberts is now only suggesting 100% annual growth, and that is also the rate underlying the recent Nemertes Research study . Others, such as the Cisco white paper , project growth rates to continue around 50% per year. Even the recent Gilder-Swanson study warning of the dangers of the exaflood, is actually assuming just a 54% annual growth rate. Now annual traffic growth rates of 50%, when combined with cost declines on the order of 33%, result in no net increase in costs to provide the increased transmission capacity. In a competitive environment, that means no increase in revenues, hardly a cheering prospect for the industry. If traffic growth could be pushed back towards 100%, where it used to be for many years, we would have a pressure for increased revenues, and also for new technologies (whether Larry Roberts' flow-aware routing, or others). And that would be a much nicer environment for the industry, and for the economy as a whole, with greater opportunities and needs for innovation. So a tentative conclusion is that the industry should worry more about inducing higher growth rates of Internet traffic, and less about imposing limits. That there are grounds for concern is shown by the example of Hong Kong (see for links to detailed government-collected statistics). Traffic levels there are about 6x those of the U.S., and the speeds of residential connections make those in the U.S. seem laughable. Yet traffic growth in Hong Kong has decelerated, and at the end of 2007 was down to almost 20% per year. Such rates, far below rates of progress in transmission technology, should strike fear in the hearts and minds of telecom professionals. Yet another conclusion is that we need more solid data on Internet traffic. The secrecy that led to the perpetuation of the "Internet doubling every 100 days" myth and the resulting collapse is still a threat, even if a less severe one. There are already a few government and cooperative industry data collection efforts, with pointers at , which have not results in any visible harm to any participants. Better information sharing about what the real traffic levels are would be healthy for service providers and their vendors, and would let us decide for certain whether we are being menaced by too much traffic or too little.