CDN Latency Benchmark

CDN latency benchmarks evaluate how effectively a content delivery network serves assets close to users. A CDN caches content at globally distributed edge locations so that requests are answered near the client rather than at a distant origin. The benchmark measures the latency and cache effectiveness this produces across regions and request types.

Good CDN performance is about both proximity and hit ratio. An edge that is geographically close but frequently misses the cache must still pay the round trip to origin, erasing the benefit.

What It Measures

Primary metrics are edge latency and time-to-first-byte (TTFB) measured from many vantage points, cache hit ratio, and origin offload (fraction of requests the origin never sees). Benchmarks also report tail latency by geography, performance for static versus dynamic content, and TLS handshake overhead at the edge.

Methodology

Distributed probes around the world request a representative set of assets and record DNS resolution, connection setup, TLS handshake, and TTFB. To measure cache behavior fairly, tests separate cold (first-fetch) and warm (cached) requests and report hit ratio over a realistic request distribution. Origin offload is computed by comparing edge request volume to requests that reach the origin. Tests vary asset size, cacheability headers, and content type, and they probe from both well-connected and poorly-connected regions to expose worst-case geography. Repeating measurements over time captures variability from routing changes and edge warm-up.

How to Interpret Results

Low median latency from rich regions is easy; the meaningful signal is tail latency from underserved geographies and the cache hit ratio. A high hit ratio with strong origin offload indicates the CDN and cache policy are well configured. If TTFB is high despite a nearby edge, suspect cache misses, slow origin, or chatty TLS. Compare static and dynamic results separately, since dynamic or personalized content benefits less from caching and more from edge compute and connection reuse. Geographic spread of latency is often more important than the global average for user experience.

Limitations

Results depend on probe placement and network conditions, so a benchmark can reflect the measurement network more than the CDN. Cache hit ratios are sensitive to the request mix and warm-up state and are easily gamed by repeatedly fetching one asset. Real user traffic patterns, cookie and query-string variation, and personalization defeat naive caching in ways synthetic tests miss. Provider edge footprints and peering change continuously, so point-in-time comparisons should be refreshed and treated as directional.