ResoNet: Resilient Optimized Network Synthesis

Internet networks are essential to our society.

Yet, recent trends have shown a growing friction between the assumptions made decades ago and the unprecedented scale at which today´s networks operate. Global electricity consumption of datacenter networks are expected to rise from 1% to 20% by 2030.

To reduce energy usage, operators must today accurately capture low-level details of all the devices in a network and their traffic characteristics.

As an example, the throughput of a network card may drop by 90% depending on the incoming traffic.Network synthesis is a recent methodology adopted by large-scale networks to produce correct-by-design per-device configurations.

However, none of the existing works can guarantee performance properties as synthesizers lack low-level details about the network devices.

A key obstacle towards performance-aware network synthesis is the lack of tools to accurately predict the performance of a network system from its high-level specification.In this proposal, called ResoNet, we propose an ambitious research agenda that aims at developing new network synthesis methods that guarantee performance and robustness requirements.

We will empirically derive accurate models of the network that are used by a formal framework for transforming a high-level network specification into a semantically equivalent one while optimizing performance. Based on our preliminary results, this may reduce energy usage of the datacenter servers by a factor of 3x.