Today’s Internet is no longer fit for the user traffic patterns that it is serving. To remedy this cognitive mismatch between the service platform and the traffic it serves, several future Internet architectures such as Content-Centric Networking (CCN) have been proposed recently with the aim of re-engineering the Internet towards supporting content-oriented communication. With the mechanisms to transform the Internet from being host-centric to becoming content-centric well spelt-out in the CCN standard, the problem of how to manage congestion and control traffic flows in CCN is still left open. Existing congestion control mechanisms for the current Internet have been shown to be ill-suited for CCN making the case for a clean-slate design of congestion control and traffic manage...[ Read more ]

Today’s Internet is no longer fit for the user traffic patterns that it is serving. To remedy this cognitive mismatch between the service platform and the traffic it serves, several future Internet architectures such as Content-Centric Networking (CCN) have been proposed recently with the aim of re-engineering the Internet towards supporting content-oriented communication. With the mechanisms to transform the Internet from being host-centric to becoming content-centric well spelt-out in the CCN standard, the problem of how to manage congestion and control traffic flows in CCN is still left open. Existing congestion control mechanisms for the current Internet have been shown to be ill-suited for CCN making the case for a clean-slate design of congestion control and traffic management mechanisms for CCN.

In this thesis, we identify the fact that congestion in CCN can take place not only in the transmission buffer but also in the pending interest table (PIT), a data structure that keeps track of all requests received from downstream nodes and forwarded to upstream nodes. Keeping this in mind, we make three contributions in this thesis: First, we characterize the PIT occupancy distribution using a 2-dimensional continuous-time Markov chain model to study the impact of PIT entry timeout and interest retransmission on the interest blocking probability. Second, given the dependence of the PIT occupancy on the PIT entry timeout and interest retransmission, we investigate the performance of two types of routers in lossy networks: no-rtx routers that do not retransmit pending interests upon timeout, and rtx routers that do retransmit pending interests periodically. Based on this, we further introduce a novel adaptive method to estimate the PIT entry timer that relies on the data chunk response delays to replace the currently used fixed-value method introduced in CCN. Finally, identifying that content requesters should be responsible for retransmitting timeout requests and that estimates of the retransmission timeout should reflect the network load conditions, we propose a novel congestion control mechanism for CCN that takes into account both the PIT and the transmit buffer. Using the PIT occupancy as a good estimator for the data flight size to arrive to the node in the near future, we design a congestion avoidance mechanism that adjusts the request rate based on anticipated congestion.