The Internet Is Full of Zombies and Nobody Is Talking About It

The speaker introduces the concept of "BGP zombies" as a hidden phenomenon affecting the internet's infrastructure. BGP, or Border Gateway Protocol, is defined as the system that manages the global map of IP addresses and traffic delivery. The speaker explains that while this system usually runs smoothly, these "zombies" represent unexpected glitches in how networks communicate. This section establishes the fundamental role of BGP in keeping the global internet connected and functional. It sets the stage for a deeper technical dive into why these routing errors occur.

In the BGP system, networks advertise new routes to attract traffic and withdraw old ones when a path is no longer valid. This process is standard for maintenance, moving customers to new edges, or shifting traffic between data centers. However, a BGP zombie occurs when a route is officially withdrawn but some networks continue to believe it exists. These outdated routes refuse to disappear from the global routing table despite the original network removing them. This mismatch between the actual network state and the routing table is the core definition of the zombie problem.

When traffic hits a BGP zombie, it fails to reach its intended destination because the path no longer physically exists. The traffic might loop between routers indefinitely or take long, inefficient detours across multiple networks until it hits a dead end. For the end user, this manifests as hanging web pages, connection timeouts, or apps failing to load for brief periods. While some instances are barely noticeable, others cause significant and visible slowdowns in internet performance. Understanding this impact is crucial for network engineers who aim to maintain high availability and low latency.

The delay in updating the global map stems from how BGP processes changes when a specific route disappears. Routers must search for a less specific fallback route, and during this computational window, some fail to clear the old, stale entry. Cloudflare's research indicates that these zombies typically persist for 6 to 11 minutes in large-scale networks. Interestingly, IPv4 zombies tend to survive longer than their IPv6 counterparts due to various technical factors. Eventually, the system self-corrects through BGP timers and ongoing communication between neighboring routers that forces a consensus.

Cloudflare discovered this issue while managing Bring Your Own IP (BYOIP) customers, where clean handoffs are essential for traffic migration. To solve the problem, they moved away from simple withdrawals and adopted a more stable two-step announcement method. By first announcing a route from a stable location, they force surrounding routers to switch to the new path before the old one is removed. This prevents the fallback search that triggers zombies and ensures a much smoother transition for global traffic. The video concludes by emphasizing that while the internet is a complex coordination of millions of decisions, such technical hurdles can be overcome with smart engineering.