This post tells the story of how my AWS colleagues put theory into practice to build a flat data center network at scale. The post provides a detailed overview of an even more detailed academic paper (RNG: Flat Datacenter Networks at Scale). Instead of the traditional network topology which stacks routers in a hierarchical, org-chart-like structure, this model connects them in a flat structure guided by randomness. This proved to be faster, more resilient, and more cost-effective but three problems had to be solved: Connecting millions of randomly assigned fiber optics cables without creating an unmanageable tangle. Routing data through a network that has no fixed structure. Proving that it would function as desired before committing time and money to build it. There's a lot of cool info in the linked post, including how they built custom hardware to shuffle connections, and how they used 530 compute years of EC2 time to test against hundreds of thousands of failure scenarios.…