This paper explores the use of multi-agent deep learning as well as learning to cooperate principles to meet strict service level agreements, in terms of throughput and end-to-end delay, for a set of classified network flows. We consider agents built on top of a weighted fair queuing algorithm that continuously set weights for three flow groups: gold, silver, and bronze. We rely on a novel graph-convolution based, multi-agent reinforcement learning approach known as DGN. As benchmarks, we propose centralized and distributed deep Q-network algorithms and evaluate their performances in different network, traffic, and routing scenarios, highlighting both the effectiveness of our proposals and the importance of agent cooperation. We show that our DGN-based approach meets stringent throughput and delay requirements across different scenarios, decreasing silver and bronze flow median waiting delays by more than 50 % and reducing the SLA violations of the latter by nearly 60 %, with respect to a classic priority queuing approach.

Tactical Traffic Engineering (TE) solutions are a must to adapt traffic steering when unexpected congestions occur. While already available, centralized solutions to locally optimizes congested tunnels and links suffer from a slow reaction time of several minutes. To address this issue, we propose a distributed Congestion Mitigation (CM) mechanism that leverages Segment Routing (SR) to offload traffic away from congested links using Unequal Cost Multi Paths (UCMP) over alternative paths. In this paper, we introduce an efficient algorithm for alternative paths’ computation, and two methods to compute UCMP weights, depending on whether remote link loads are available or not. We show that the proposed path computation method is faster than a modified K-shortest path algorithm. For traffic splitting, we show that the knowledge of remote link loads and per-destination traffic is a key to mitigate congestions in loaded scenarios, approaching the results obtained with an optimal solution. However, when not available, a local solution can already mitigate congestions in lightly loaded scenarios.

Network automation in the fifth generation of mobile networks (5G) requires network slices to be efficiently computed and deployed. In order to guarantee physical isolation for virtual networks and no interference between different slices, it is possible to rely on hard slicing, whose principles can be implemented using Flex Ethernet (FlexE) technology. As slices are created and deleted over time, it is necessary from time to time to defragment resources and reoptimize bandwidth reservations for the remaining slices.In this demo, we present our algorithmic framework, based on Column Generation, and we showcase how to efficiently defragment the network in order to reduce the Maximum Link Utilization (MLU) of current reservations with a minimum number of configuration changes.