Delay-tolerant networking: an approach to interplanetary Internet

In this paper, a variety of network scientists and researchers (including course guest speaker Vint Cerf) propose a network architecture that would be suitable for an interplanetary Internet. The authors begin by identifying a variety of differences between electronic communication through space and on Earth, including: higher latency and lower data rates in space than on Earth, lack of consistent end-to-end connectivity that is characteristic of Earth networks, and much lower potential for congestion on space networks given they are likely to be centrally managed. The Internet protocols (IP) are not equipped to handle these differences. In particular, TCP’s reliance on an end to end connection (for ensuring data acknowledgement) will result in too many delays since the many-hop system in the middle will not be efficiently used. UDP is likewise unsatisfactory because there is no data acknowledgement at all. These difficulties only get worse as other routing protocols like BGP are built upon TCP and suffer from the same issues.

To solve these challenges, the authors propose a delay-tolerant network (DTN) architecture characterized by three fundamental principles: 1) postal model of communication: since latency can be “arbitrarily long,” communication is asynchronous and all necessary information (e.g. metadata, authentication information, etc.) for a particular request is collected into a bundle and sent together; 2) tiered functionality: each link may use whatever protocol is optimal (e.g. bundling protocol described in 1, TCP, etc.); and 3) terseness: assume bandwidth is at a premium so optimize for small size at the cost of increased processing. In reviewing the proposal, the authors do recognize an unsolved security challenge. A back-and-forth key exchange would likely take too long over the high latency network, so certificates may need to be sent with each bundle, though that may violate the terseness principle since those can be large.