This thesis addresses the previously unanswered question of how to handle best-effort traffic in Multiprotocol Label Switched (MPLS) based next generation data-centric networks. Specifically, this thesis examines the technical feasibility and assesses the performance of implementing an integrated framework for managing and traffic engineering best effort traffic in next generation data centric networks. Through proof and simulation, this thesis will show that the proposed integrated framework successfully enforces weighted fair rates for best-effort flows while simultaneously increasing.

The implementation of the proposed system proceeded in two phases. In the initial phase, a system for signaling fair rate using unmodified RSVP messages is devised. Signaling overhead and legacy integration are analyzed. To calculate fair rates, a distributed WPMM fair rate algorithm was developed. The algorithm is tuned to work at low refresh rates and calculates weighted fair rates with O(1) computation complexity. An elegant proof based on bottleneck ordering is provided to show that the algorithm converges to fair rates in arbitrary network topologies.

Based on the results of phase one, a second phase was undertaken to introduce a new paradigm in queue management. The introduction of this new queue management was essential to ensure that the initial congestion control mechanism is compatible/friendly with the native TCP congestion control mechanism, and will converge to fair rates with minimal signaling overhead. This new paradigm is based on Better than Best Effort (BBE) and Less than Best Effort (LBE) fair rate marking of packets and is shown to drive the network to fair rates with minimal overhead. This mechanism is shown to be vastly superior to the initial mechanism in both enforcing fair rates and in network throughput. We call this system TERM (TCP friendly Explicit Rate congestion control for MPLS networks).

Low frequency signaling coupled with efficient calculation of WPMM fair rates makes the TERM system a feasible solution for explicit rate congestion control in large scale networks. To our knowledge this is the first implementation of such a system.