Prediction-based Flow Routing in Programmable Networks with P4
Christoph Hardegen, Sebastian Rieger · 2020
Distributing the load of arising network flows with varying characteristics across available topology links and paths efficiently and up to capacity is challenging. To allow for an optimized (e.g., equalized) utilization of multiple links or paths, a flow routing paradigm and architecture leveraging topology state snapshots, predicted flow characteristics and data plane programmability is proposed. This flow steering strategy is compared to existing shortest and multi-path routing methods. Moreover, enhancements to increase the practical applicability of the proposed solution are discussed.