Whiteboard exercise. Try the problem cold, then reveal the rubric to self-score.
Out of 10 points45 min whiteboardReference solution →
01
Prompt
A global Content Delivery Network: serve cached copies of customer content from hundreds of edge points-of-presence (PoPs), delivering to users with sub-50ms RTT anywhere on Earth. The hard parts: global routing so requests land at the closest healthy PoP (Anycast); a cache hierarchy that protects origin from stampedes at 10M+ req/sec; and cache invalidation that reliably purges stale content from ~300 locations in seconds, not hours. Cloudflare: 300+ PoPs, ~55M HTTP req/sec, serves ~20% of internet traffic. Akamai: 4000+ PoPs, older + deeper model.
Time budget: 45 min whiteboard. Draw architecture, estimate numbers, discuss tradeoffs.
02
Hints (progressive — click to reveal)
Hint 1
Anycast first, not DNS. Modern CDN answer. Explain the mechanism in one sentence.
Hint 2
Tiered cache beats flat-edge. Without it, your origin burns. Mention by name; it's a load-reducing design choice that shows sophistication.
Hint 3
Purge is not magic. It's pub/sub with seconds of propagation. Don't say "instant cache invalidation."
03
Rubric — 10 points
+2 Anycast first, not DNS. Modern CDN answer. Explain the mechanism in one sentence.
+2 Tiered cache beats flat-edge. Without it, your origin burns. Mention by name; it's a load-reducing design choice that shows sophistication.
+2 Purge is not magic. It's pub/sub with seconds of propagation. Don't say "instant cache invalidation."
+2 Request coalescing for stampedes. One-request-in-flight per key. Classic pattern that many candidates forget.
+1 Analytics + logs. CDN's real product is observability. Mention shipping edge logs → central analytics; it's half the value customers pay for.
+1 Stale-while-revalidate. One line, massive impact. Shows you know the production-grade answer.
Self-score: tally the points you would have mentioned unprompted. 7+ is interview-ready on this problem.
04
Red flags (things that tank the interview)
No back-of-envelope estimation — jumps straight into components without quantifying scale for CDN
Single point of failure — no replication, failover, or redundancy discussed
Ignores data model and storage choices — hand-waves the database layer