Worldcom's current financial troubles are raising concern about a possible shutdown of Worldcom's ISP services. Although this may seem farfetched, similar problems resulted in the shutdown of much of KPNQwest's network in Europe earlier this year. Assuming a complete shutdown of the UUNet backbone (comprising AS701, AS702, and AS703), how much of the Internet would this affect?

We used a BGP routing table snapshot from June 26, 2002 to estimate how much of the Internet depends on the UUNet backbone. This table was supplied by the University of Oregon's Route Views project, and includes nearly 50 complete views from the default-free zone of the Internet backbone. PCH maintains an archive of daily snapshots of these BGP tables, and the snapshot used for this analysis is available from the PCH archive as 2002-06-26_03:24:00_PDT.gz

Our estimate of the number of hosts connected to the Internet comes from the Internet Domain Survey work done by Network Wizards. To estimate the number of hosts using the UUNet backbone, we calculated the average number of hosts per reachable IP address for the Internet as a whole. We then multiplied this by the number of reachable IP addresses in each category to obtain a rough estimate of how many hosts would be affected by a complete shutdown at Worldcom.

The first row of this table shows our baseline numbers for the Internet as a whole. The second row includes all hosts whose `best routes' from either Sprint (AS1239) or Cable & Wireless (AS3561) pass through UUNet's backbone. The third row includes all hosts whose route originates with UUNet, or whose origin AS is single-homed to UUNet's backbone. The last row includes all hosts who are single-homed to UUNet's backbone even though their origin AS does not exchange routes directly with UUNet.

The most obvious conclusion from this analysis is that ~10% of the hosts connected to the Internet would be disconnected by a complete Worldcom shutdown. However, virtually all of these have a direct business relationship with Worldcom, and should already be aware of their dependence on the company for Internet access. Only a very small number of hosts (~0.5% of the total) may not know that they are dependent upon UUNet for their Internet connectivity. These customers would probably be very surprised to find themselves cut off.

This situation demonstrates the clear advantage of purchasing transit at an exchange point, as opposed to paying for one or more private connections to multiple backbone providers' networks. In the event of major backbone provider's failure, customers can much more quickly reestablish their connectivity by purchasing transit from another backbone provider at the same exchange point.