Mistill: Distilling Distributed Network Protocols From Examples

Abstract

Traffic Engineering (TE) mechanisms in data center networks make distributed forwarding decisions based on the global network state. Thus, new TE mechanisms require the design and implementation of effective information exchange and efficient decentralized algorithms to compute forwarding decisions, which is challenging and time-intensive. To automate and simplify this process, we propose Mistill. Mistill distills the forwarding behavior of TE policies from exemplary forwarding decisions into a Neural Network. Mistill learns (i) how to encode local state into update messages, (ii) which network devices must exchange updates, and (iii) how to map the exchanged updates into forwarding decisions. We demonstrate the abilities of Mistill by learning three TE policies, verifying their performance in simulations on synthetic and realworld traffic patterns, and by showing that the learned policies generalize to unseen traffic patterns. We implement Mistill as a proof-of-concept and show that Mistill reacts on average within 1.3 ms to changes in the network.