Content Delivery and Edge Networking Questions

Situation & goal
I would guide the team to choose the caching tier (client vs edge CDN) by scoring each option on criteria that matter to product, reliability, and ops, then validating with experiments.

Evaluation criteria (definitions)

• Cache invalidation complexity — how hard are correctness guarantees when content changes (propagation latency, tools/APIs required, risk of stale-serve bugs).
• Staleness tolerance — business tolerance for serving slightly older data (per endpoint/feature).
• Operational ownership — who runs/monitors the cache, SLAs, runbook complexity, cost to operate.
• Cache hit amplification — expected effective reduction in origin load (client-local hits, CDN global caching benefits, cold-start amplification).
• Additional: latency improvement (p95 global), cost delta, security/privacy constraints.

Scoring approach

• Assign weights (example: latency 30%, staleness 25%, invalidation 15%, ops 15%, amplification 15%). Score 1–5 for each option per criterion. Compute weighted sum to rank choices. Make weights configurable per product area.

Validation experiments

• Run A/B or regional rollout for a week:
  • Measure p50/p95 latency, origin QPS, cache hit ratio, error/revalidate rates, and staleness incidents.
  • Implement synthetic invalidations to test worst-case propagation and measure time-to-consistency.
  • Track operational overhead (alerts, pager counts) and cost.
• Stop criteria: missed SLOs, unacceptable staleness incidents, or cost blowouts.
• Use findings to adjust weights and iterate.

As an engineering manager I’d involve product, SRE, and frontend teams in tradeoff weighting, own the rollout plan and fallout runbooks, and ensure learnings become documented standards.