# Network

request patterns, caching, retries, connection reuse.

## Measure

- network tracing → request/response patterns
- monitoring → latency, throughput, error rates
- connection profiling → establish, reuse, teardown
- metrics → p50/p95/p99 latency, req/sec, error rate, pool utilization

## Optimize

- connection pooling → reuse instead of new
- caching → client, server, edge
- batching → combine requests into single round-trip
- compression → payload size reduction
- edge caching → serve from closer locations
- retry tuning → avoid thundering herds, excess retries

## Pitfalls

- over-caching → stale data, invalidation overhead
- geography → caching helps, distance still matters
- pool misconfig → too few queues, too many wastes
- server latency ≠ total; propagation delay counts