Internet traffic bypassing the US?

Posted on August 31, 2008  /  3 Comments

Ten years ago, pretty much all the traffic went through the US Internet backbone. Today, claims are being made that only 25 per cent of traffic is routed through the US system.

This may require changes in LIRNEasia’s (and Singapore’s) efforts to improve broadband quality of service experience through benchmark regulation or otherwise, using as one of the measures, Round Trip Time to the Internet cloud, defined as first point of landing in the US.  An alternative will not be easy to come by, but we have faith in the wisdom of the many.   Please contribute.


  1. Reduction in US bypass traffic can be largely due to the growing Asian Internet user base. Imagine China. Vast majority of Chinese users obviously use content in Mandarin and has no reason to access US servers. In addition, mirror servers further cut down international traffic.

    However, it is not the case everywhere. In the absence of useful content locally and the lack of mirror servers force the users in third world countries to reach US servers. So international bandwidth continue to be an issue for them.

    So benchmarking RTT for the first foreign entry point is still valid for many countries. May be even for Singapore.

    Finally it is the ‘first entry point to a foreign ISP’ – and not necessarily the first entry point to USA. (depending upon the point the local ISP connects) For example, for Sri Lanka, it can be first entry point to Singapore.

  2. How can first entry point in Singapore work? Your research showed that there are considerable delays from Singapore to US. The local provider buys the capacity from Singapore to the US and is thus responsible.

    First entry point to the cloud is the key. If one enters the cloud in Singapore, yes. But that has to be established first.

  3. The ‘cloud’ has started to shown increasingly ‘non-cloud’ characteristics, adding to the complexity of the problem.

    So a more logical approach will be to measure RTT to every node in the way and take it case by case – which will be done at the next phase. The software notes down RTTs to every node in the way. This data will be useful to find the right reference first entry point.

    Also RTT is only one measurement of quality. One drawback of RTT is unlike other metrics it depends on the distance packets travelled, so a satellite link gives a higher RTT than a terrestrial one, even if there is enough bandwidth. Non-interactive applications (eg. Video on Youtube) can tolerate fairly high RTTs if the remaining parameters stay intact.