It is generally accepted that some of the architectural principles of the Internet should be revised to better address the requirements of today’s dominating applications and traffic types that are fundamentally different from those at the time of the original design. In 2008, the Network Architecture Project will focus on fundamental research for offering architectural solutions to the routing problems created by mobility of the end points, prevention of distributed denial of service (DDoS) attacks, and transport protocols addressing the needs of real-time media delivery.

The current Internet was originally designed for static computers. One key consequence of this static node assumption is that IP addresses were used both as identity and location. This paradigm is not sustainable once the communication endpoints move about. As a consequence, the advent of mobile telecommunications and mobile computing will require significant modifications to the current architecture. We believe that a clean slate effort at designing a new routing and addressing architecture which efficiently supports mobility will yield valuable lessons to be integrated in future Internet architecture. As an example, one of our key focus areas is flat identifier routing, a novel kind of architecture which uses a distributed hash table (DHT)-like functionality to route packets from source to destination. We intend to study the fundamental performance of such architectures, while proposing ways of making them manageable from a network operator perspective.

In recent years, DDoS attacks have evolved from malicious hacker exploits into wide scale organized crime activities and government-sponsored attacks and became a number one security threat. We aim to create a new network architecture that provides a high level of robustness against DoS and DDoS attacks. We argue that one of the main downsides of the current Internet architecture when it comes to cost distribution is that the cost lies at the receiver side and no other participants have incentive to defend against attacks. In our approach we propose a new architecture that rearranges the economic incentives to remove burden of dealing with unwanted traffic from the receiver and distributes the cost of attacks among all participants. Consequently, this approach significantly increases the cost of the attack for the adversary and provides incentive for all the users and the Internet Service Providers (ISPs) to collaborate and participate in network defense.

TCP/IP has been mostly successful and sufficient for static Internet with dominating HTTP traffic. With mobility being more widespread and payloads of real-time traffic becoming the dominant payload, there is a need to question the original assumptions of TCP, TCP fairness, and TCP friendliness arguments. One of our goals is to create a synergy between forward error correction (FEC) schemes and congestion control for media delivery, while reversing the argument of TCP friendliness with media-friendliness. Another goal of the project is to look at the transport layer redesign in conjunction with the overall network architecture design in a clean-slate fashion. Our overarching approach is instead of patching the current architecture with a proxy-based infrastructure to make TCP work as the architecture is evolved, we should first determine the desired features (e.g., mobility, naming, scalability, etc.) we want in the network architecture to deliver a multitude of services and then we should design an appropriate transport layer that works seamlessly over these desired features.

The networking research community is in agreement that virtualization in the network forwarding engines, like virtualization in operating systems, is expected to offer significant benefits for diversifying routing and addressing architectures and simplifying the migration of successful services into the network. Virtualized networks are therefore expected to offer an excellent environment for rapid implementation and testing of some of the architectural ideas created in this project, and NML expects to prototype our ideas on the coming prototype deployments of virtualized networks in addition to extensive software simulations, experiments on overlay testbeds, such as PlanetLab, and emulation testbeds, such as EmuLab.