A data-oriented network architecture

Abstract

In the 25 years since becoming commercially available, the Internet has grown into a global communication infrastructure connecting a significant part of mankind and has become an important part of modern society. Its impressive growth has been fostered by innovative applications, many of which were completely unforeseen by the Internet's inventors. While fully acknowledging ingenuity and creativity of application designers, it is equally impressive how little the core architecture of the Internet has evolved during this time. However, the ever evolving applications and growing importance of the Internet have resulted in increasing discordance between the Internet's current use and its original design. In this thesis, we focus on four sources of discomfort caused by this divergence. First, the Internet was developed around host-to-host applications, such as telnet and ftp, but the vast majority of its current usage is service access and data retrieval. Second, while the freedom to connect from any host to any other host was a major factor behind the success of the Internet, it provides little protection for connected hosts today. As a result, distributed denial of service attacks against Internet services have become a common nuisance, and are difficult to resolve within the current architecture. Third, Internet connectivity is becoming nearly ubiquitous and reaches increasingly often mobile devices. Moreover, connectivity is expected to extend its reach to even most extreme places. Hence, applications' view to network has changed radically; it's commonplace that they are offered intermittent connectivity at best and required to be smart enough to use heterogeneous network technologies. Finally, modern networks deploy so-called middleboxes both to improve performance and provide protection. However, when doing so, the middleboxes have to impose themselves between the communication end-points, which is against the design principles of the original Internet and a source of complications both for the management of networks and design of application protocols. In this thesis, we design a clean-slate network architecture that is a better fit with the current use of the Internet. We present a name resolution system based on name-based routing. It matches with the service access and data retrieval oriented usage of the Internet, and takes the network imposed middleboxes properly into account. We then propose modest addressing-related changes to the network layer as a remedy for the denial of service attacks. Finally, we take steps towards a data-oriented communications API that provides better decoupling for applications from the network stack than the original Sockets API does. The improved decoupling both simplifies applications and allows them to be unaffected by evolving network technologies: in this architecture, coping with intermittent connectivity and heterogenous network technologies is a burden of the network stack.