One of the big challenges facing the Internet is the rapid evolution of both the technologies and services that are being deployed, in addition to an increase in user expectations with respect to quality of experience.  The reason is that core data and control protocols, such as IP and BGP, were developed in the 70?s and 80?s and evolving these protocols is challenging since they need to interoperate across the entire Internet.  In addition, optimizing quality of experience requires information about resource availability, which was not considered in the design of the core Internet protocols.

To meet the rapidly evolution in the Internet infrastructure, services, and user expectation, we identified three requirements for the Internet control plane: First, control protocols must be evolvable so they can adapt to future emerging requirements. Second Internet routing protocols must be resource aware. Finally, the Internet control plane must support coordination with the control planes of increasingly sophisticated edge networks such as CDNs and cloud infrastructures, which is critical for optimizing the quality of service delivered to users.

We considered the issue of evolvability in three separate projects.  First, we developed techniques that allow the co-existence of multiple routing protocols in the Internet. We also develop tools to verify the correctness of control plane, which is important when control protocols are updated more rapidly.  Finally, we developed a tool that allows us to analyze the interactions of diverse congestion control protocols which have become increasingly diverse.

We also explored different ways of supporting resource awareness.  A first solution uses resource brokers to establish fixed-bandwidth pipes between content-providers and stub-networks, allowing content providers to coordinate resource allocation in their data centers and across the Internet.  We develop ``data-driven?? networking techniques that use continuous measurements to inform service providers about available network resources.  Finally, developed efficient sketch-based algorithms that collect diverse traffic statistics that can be used by various control planes to optimize resource allocation.

Finally, we explored the issue of coordination across control planes in a several contexts since the solutions depend on the type of resources involved and the application domain.  For example, a challenge in many domain is that resource allocation decisions made by one control plane can impact other control planes both in terms the decision they make and the outcomes.  We developed techniques to explicitly coordinate, e.g., by using a broker for video in an internet context, or by explicit notification messages in hybrid optical-electrical data center networks.  Another approach  is to allow one control plane to delegate control of a set of its resource to another control plane, effectively isolating the decisions made by the two control planes.  We also developed solutions for wireless access networks.  Examples include access point selection, which depends on the service quality different operators can provide, and computational offloading techniques that must adapt to the available bandwidth on the access link.  Finally we studied how to safely delegate control of network functions to a Managed Service Provider (MSP) based on the least privilege principle.

Last Modified: 03/02/2023
Modified by: Peter A Steenkiste