A provider evaluation should end with a tested operating model: how you publish, secure, observe, and recover the delivery path.

Network

Ask for evidence about regional reach, congestion, throughput, and tail latency. Confirm with your own workload from representative access networks; a network-wide average is not a result for your application.

The delivery model

How it works

What to look for in a CDN provider

  1. 01ProveUse your own representative content and regions.
  2. 02OperateTest controls, logs, automation, and access.
  3. 03RecoverRehearse failure and a move away.
Conceptual model for this configuration. The actual path depends on provider policy and application behavior.

Miss path and media

Origin Shield, origin groups, image transformation, live and VOD, large files. If your workload is any of those, a static-only edge will show up on the first event.

Operate it

Purge, prefetch, logs, API, managed and custom TLS. If those are tickets, you do not operate the edge.

Leave production DNS for last

Use a test hostname to validate delivery, TLS, and application behavior before changing production DNS. A separate DNS migration is optional and should have its own preparation and rollback plan.

Validate before production

Document the evidence for each requirement, including API behavior and plan limits. Run a controlled invalidation and origin-failure test. Confirm data export, hostname ownership, and a rollback procedure before making the trial production-critical.

  • Record the configuration and expected result before changing traffic.
  • Keep evidence of the test and a working rollback for the previous behavior.

Sources & further reading