CDN tuning for a live platform

The situation

I tuned the CDN for a high-traffic platform in e-commerce that served users across time zones while it stayed live. A Content Delivery Network, or CDN, is a network of servers that stores copies of files closer to visitors. The platform already used a CDN, but pages slowed during peaks and some updates took too long to appear.

Constraints

The platform was live all day, so I could not schedule downtime. Some pages were personalized and could not be cached. Marketing links added many query strings that confused caching. The origin servers, the primary web servers, were healthy but could not absorb sudden bursts.

My approach

I made a simple plan, measured first, then changed one thing at a time. I kept a rollback for every rule. Here is what I changed and why.

• Mapped content types. HTML stayed uncacheable. Images, scripts, and styles got long cache times with versioned file names.
• Set clear headers. I used Cache-Control, ETag, and Last-Modified to guide caching and rechecks.
• Tuned the cache key, the rule the CDN uses to decide if requests are the same, ignoring known marketing parameters, stripping cookies on static files, and keeping query strings on API calls.
• Enabled Brotli and gzip for text and kept connections open. Supported HTTP 2 and HTTP 3.
• Added stale-while-revalidate and stale-if-error so cached content served fast while the CDN refreshed, and stale content served during brief origin errors.
• Used an origin shield, an extra layer inside the CDN, to funnel requests through one CDN layer and cut origin load during spikes.
• Set a safe purge routine. Purged by tag or path, never the whole cache, and prewarmed key pages before announcements.
• Added WAF rules, a web application firewall, to throttle abuse and reduce bursts from single sources.

What I monitored

• Time to first byte, the delay before the first byte arrives, from several locations.
• Cache hit and miss patterns by path groups, watching for unexpected misses.
• Spikes in server errors and slow origin responses.
• Certificate renewals, DNS changes, and CDN purge events.
• WAF blocks and any false positives that could affect real users.

What I learned

Small, careful changes kept the site steady while I tuned. Good cache keys, clear headers, a safe purge plan, and serving stale content during brief trouble prevented surprises.

If you want help, I offer Hosting & Infrastructure Plans, Web Hosting, Maintenance Plans, Website Updates, WAF & Firewall, and AppWT Analytics.