Why is my traffic going through a different node?

Our anycast network structure uses the Border Gateway Protocol (BGP) to distribute traffic to various upstream providers around the world. These providers range from Tier 1 such as Cogent, NTT, GTT and others, to smaller tiers restricted to specific continents or countries. Traffic is routed through Internet nodes, either through their route servers or by direct peering.

By default, a BGP connection is a best-effort relationship. Once the traffic leaves our site, full control lies with the upstream providers. Some of these providers have predefined agreements that affect the path traffic takes after it leaves our network.

Since we lose control of the traffic after it is forwarded to the upstream providers, there is a possibility that DNS requests will be forwarded to nodes that are far away, both in terms of network connectivity and geographic distance. This can result in, for example, a DNS request from a customer in North America being forwarded to a node in South Africa when the request should have been directed to one of the many nodes in North America.

Although many providers adjust their internal routing configuration when we point out problems, there are also large market players like Hurricane Electric that do not. Although we have ended our direct peering with them, many of our upstream providers are using them as one of their own upstream connectivity points.

Our DNS team continuously monitors these issues and contacts the providers when possible. In most cases, these issues can be resolved, but it takes time. Occasionally, we also use BGP communities to influence how our upstreams forward traffic.