How do globally distributed databases achieve low-latency writes?

Global databases (CockroachDB, Spanner, YugabyteDB) face latency challenge: consensus requires round-trips across regions. Optimizations: Spanner uses TrueTime (GPS + atomic clocks) for lock-free reads. Follower reads (read from local replica at slightly older timestamp). Write locality (place data near writers using geo-partitioning). Multi-region Paxos groups. CockroachDB uses hybrid logical clocks. Trade-offs: consistency vs latency. Consider: data placement, read/write patterns, consistency requirements per use case.