Evolution of Meta's Edge and Backbone Networks

This abstract explores the multifaceted evolution of Meta's network architecture—specifically the Backbone and Edge—in response to unprecedented traffic demands and the emergence of latency-sensitive AI workloads.

The presentation first details the scaling journey of the Backbone network, which has achieved over 10X growth through a strategic combination of DC Metro Architecture, platform Scaling Up (e.g., faster interfaces), and Scaling Out (e.g., multiple devices per plane). A key enabler of this massive scale and efficiency is the deployment of IP and Optical Integration using ZR technology, resulting in significant power and space savings within our network sites.

Next, we examine the transformation of the Edge network, driven by a shifting traffic profile heavily influenced by new AI use-cases. To meet the demand for lower-latency experiences and uncacheable, unique content generation, Meta is building an Elastic Edge that strategically moves compute resources, including a global GPU footprint, closer to end-users. This new compute locality is crucial for optimizing user experience and allows for a more efficient use of our infrastructure by reducing costly backbone traffic.

Finally, we discuss updates to our Peering strategy, including a move away from Route-Server Peering and the launch of an Automated Peering Portal. We clarify that this shift to automated, bilateral peering is intended to simplify and accelerate connections for thousands of smaller networks globally, but importantly, it does not constitute a change in our fundamental peering policy or our commitment to the open Internet.