Magellan: Toward Automatic Mapping from High-Level SDN Programs to Low-Level, Optimized Datapath Pipelines

Despite the emergence of pipelining as a key feature of nextgeneration SDN data-path models, there are no previous studies on how to automatically design, populate, and update datapath pipelines from completely datapath-oblivious, highlevel SDN programs. We refer to this as the pipeliningoblivious programming challenge. This paper presents Magellan, the first systematic investigation of the challenge. Including multiple novel algorithms, including a novel, efficient combination of symbolic map and flow-rule guided execution to compute compact pipeline contents, and an efficient pipeline design algorithm to automatically optimize pipeline resource usage and satisfy hardware constraints, Magellan achieves automatic design, population, and updates of datapath pipelines from completely datapath-oblivious, highlevel SDN programs. We implement Magellan and apply it to multiple high-level programs to demonstrate its benefits.