The following text is copyright 2005 by Network World, permission is hearby given for reproduction, as long as attribution is given and this notice is included.

Implications of an improving Internet

By Scott Bradner

Most of the Internet has been getting better over the last few years. In much of the world the Internet is now good enough for all but the most demanding applications. This improvement has been in the default "best effort" service and has not depended on Internet service providers (ISPs) implementing fancy quality of service (QoS) mechanisms. Paradoxically, some ISPs may see this news as a threat to their future financial health.

There are a number of research groups currently studying Internet performance. It is not easy to get good data about Internet performance, as KC Claffy details in one of her talks (http://http://www.caida.org/outreach/presentations/2005/cenic0503/). KC is the main instigator of Cooperative Association for Internet Data Analysis (CAIDA), an UCSD-based analysis and research group, is one of the best of the Internet-related research centers with an amazing program plan (http://www.caida.org/projects/progplan/progplan03.xml). The CAIDA web site (http://www.caida.org/) contains a wealth of interesting and important papers and presentations that try to help people understand this Internet thing.

CAIDA is not the only group doing research on the Internet. Some members of the physics community have been studying Internet performance for most of the last 10 years. The International Committee on Future Accelerators (ICFA) has had working groups thinking about Internet performance since at least 1997. The current iteration was formed in 2002 (http://www.slac.stanford.edu/xorg/icfa/scic-netmon/). These groups have published a series of reports on the state of Internet performance, the latest of which was published in January 2005. (http://www.slac.stanford.edu/xorg/icfa/icfa-net-paper-jan05/) I'm not sure why the physicists are studying Internet performance unless it's to figure out if they can use the Internet to deliver the (very) large datasets that their experiments produce, in any case, its very good work.

The report mostly deals with packet loss in data transmissions, with round trip times and with data throughput between the Stanford Linear Accelerator (SLAC) and testing points throughout the world. The set of countries that the testing points are located in represent 78% of the world's population and 99% of the world's Internet users.

The test results show that by the end of 2003 the packet loss rate to countries with 77% of the world's population is low enough that voice over IP (VoIP) will work very well (63.5%) or well enough to be easily understandable (13.8%). This is up from 48.8% in 2001. One example is reliability within the US -- this has gone down from a loss rate of over 10% in January 1995 to a loss rate of under 0.5% in January 2004. Improvements were also seen in reducing round trip times and increasing data transmission throughput.

These improvements were in the standard Internet "best effort" service. As Vonage and other overlay VoIP services have shown, VoIP "just works" to much of the world most of the time. You do not have to pay the carriers extra money for better service to make VoIP work well enough to be very useful. This fact may be a real threat to the financial well being of carriers that have been planning on making more money by charging extra for better quality service -- and that is most of the telco-based carriers. These carriers will be forced to try to make money selling a commodity service (unless more of them purposefully try to mess up their networks to mess up VoIP as Vonage has claimed that some already do). These carriers could be in for a rough ride.

disclaimer: Harvard claims to not be in the commodity service business but has not expressed an opinion on carriers that may be forced to be so -- thus the above is my own opinion.