Routing TCP[s]IP (Vol. 22001)

 

The fundamental problem with EGP is its inability to detect routing loops . Because there is an upper boundary on the distance EGP uses (255), you might be tempted to say that counting to infinity is at least a rudimentary loop-detection mechanism. It is, but the high limit combined with the typical Poll interval makes counting to infinity useless. Given a default Poll interval of 180 seconds, EGP peers could take almost 13 hours to count to infinity.

As a result, EGP must be run on an engineered loop-free topology. Although that was not a problem in 1983, when EGP was intended merely to connect stub gateways to the ARPANET backbone, the creators of EGP already foresaw that such a limited topology would soon become inadequate. The autonomous systems making up the Internet would need to evolve into a less structured mesh, in which many autonomous systems could serve as transit systems for many other autonomous systems.

With the advent of the NSFnet, the limitations of EGP became more pronounced. Not only were there now multiple backbones, but there were acceptable use policies concerning what traffic could traverse what backbone. Because EGP cannot support sophisticated policy-based routing, interim solutions had to be engineered[4].

Another major problem with EGP is its inability to adequately interact with IGPs to determine a shortest route to a network in another AS. For example, EGP distances do not reliably translate into RIP hop counts. If the EGP distance causes the hop count to exceed 15, RIP declares the network unreachable. Other shortcomings of EGP include its susceptibility to failures when attempting to convey information on a large number of networks, and its vulnerability to intentionally or unintentionally inaccurate network information.

Last but certainly not least, EGP can be mind-numbingly slow to advertise a network change. The section "Troubleshooting EGP" includes an example in which a network in an EGP-connected AS becomes unreachable. As the example demonstrates , almost an hour passes before a gateway four hops away determines that the network has gone down.

Several attempts were made to create an EGPv3, but none were successful. In the end, EGP was abandoned in favor of an entirely new inter-AS protocol, BGP. As a result, Exterior Gateway Protocol is now not only the name of a protocol, but the name of a class of protocols, giving rise to the notion of an EGP named EGP. Nonetheless, the legacy of EGP is still with us today in the form of autonomous systems and inter-AS routing.

Категории