openskills.info
Course Preview

Global Traffic Management

Global traffic management decides which healthy application location should receive a user’s connection. It combines traffic-steering policy, health signals, and global delivery mechanisms such as DNS, anycast, and proxies.

itWeb servers, proxies, and traffic management

Don't Panic — Global Traffic Management

Global traffic management decides which application location receives a new connection. It exists because putting a service in several regions does not make the internet pick the healthy, permitted, or adequately provisioned one by telepathy. Before this control layer, each client reached whatever address it had, and the application discovered the consequences later, usually at an inconvenient volume.

The useful shape is a small control loop: signals establish eligibility, a steering policy chooses among eligible destinations, and a steering plane applies that decision. The three common planes are DNS, anycast, and a global proxy. DNS changes an authoritative answer; anycast lets routing choose one location for a shared address; a proxy receives the connection and selects an origin pool. They can work together, because apparently the internet enjoys giving the same job several hats.

DNS has the first surprise. A DNS traffic manager does not handle the connection it influences. A recursive resolver, the DNS server that retrieves and caches answers for clients, can retain an answer for its time to live. A lower TTL shortens the intended cache period, but it does not recall answers already cached or turn DNS into a per-request switch. Keep an old destination available while safe cache overlap passes.

Health is evidence, not truth. A TCP check says a port accepted a connection. An HTTP check can test a path and response. Neither proves that every dependency or critical user journey works. Probe from relevant locations, use failure and recovery thresholds, and avoid a check that either misses the request path or removes every location because one shared dependency failed.

Failover is also a capacity event. In active-active placement, several locations serve traffic and must handle data and session behavior across locations. In active-passive placement, the recovery location waits and can drift. Neither arrangement manufactures spare capacity, moves sessions, or replicates data. The surviving location must already be able to accept the transfer.

Read the Course tab for the full control loop and failure behavior. Use Slides to compare the three steering planes and their timelines. Keep Cheatsheet nearby when choosing health signals, rollout weights, and the manual override that will seem unnecessary right up until it becomes the most interesting control in the room.

Where this skill leads

Relevant careers

See how this topic contributes to broader role-level skill maps.

Sources