Cloud_Infra
Entry: 002
CDN Edge
Caching_Latency
Published: 2025.02.28
Read_Time: 6 min
The Problem of Global Latency
When serving media-heavy content to a global audience, every millisecond counts. In our initial deployment, users in APAC were experiencing P99 latencies of over 1.2s due to origin servers being located in us-east-1.
Tiered Caching Strategy
We implemented a three-tier approach to maximize cache hit ratios while ensuring content freshness:
- L1 (Application/Redis): Frequently accessed metadata cached in-memory.
- L2 (CloudFront Edge): Regional caching with surgical invalidation policies.
- L3 (Shield/Origin): Protection and centralized fallback.
Technical Insight
"The trick isn't finding the perfect invalidation strategy — it's designing your system so you rarely need to invalidate at all."
Results & Impact
| Metric | Before | After |
|---|---|---|
| P50 Latency (Global) | 320ms | 45ms |
| Origin Load | 100% | 12% |