DeepConfig: Automating Data Center Network Topologies Management with Machine Learning

  • 2017-12-11 17:07:12
  • Christopher Streiffer, Huan Chen, Theophilus Benson, Asim Kadav
  • 29

Abstract

In recent years, many techniques have been developed to improve theperformance and efficiency of data center networks. While these techniquesprovide high accuracy, they are often designed using heuristics that leveragedomain-specific properties of the workload or hardware. In this vision paper, we argue that many data center networking techniques,e.g., routing, topology augmentation, energy savings, with diverse goalsactually share design and architectural similarity. We present a design fordeveloping general intermediate representations of network topologies usingdeep learning that is amenable to solving classes of data center problems. Wedevelop a framework, DeepConfig, that simplifies the processing of configuringand training deep learning agents that use the intermediate representation tolearns different tasks. To illustrate the strength of our approach, weconfigured, implemented, and evaluated a DeepConfig-Agent that tackles the datacenter topology augmentation problem. Our initial results are promising ---DeepConfig performs comparably to the optimal.

 

Quick Read (beta)

loading the full paper ...