Why Distributed Systems Can Never Truly Agree on 'Now'

Distributed systems routinely assume their clocks are synchronized, but NTP only keeps clocks within 1–100 milliseconds of each other — close enough for casual use, but unreliable for establishing event ordering across services. This clock skew means a timestamp comparison like 'A happened before B' can be outright false, not due to bugs or hardware failure, but by design. Computer scientist Leslie Lamport addressed this in 1978 by introducing logical clocks, which track causality through incrementing counters rather than relying on wall-clock time. Building on that, vector clocks allow each service to maintain per-node counters, enabling systems like DynamoDB and Riak to detect truly concurrent, unrelated events. The core lesson mirrors Einstein's special relativity: there is no universal 'now,' and reliable distributed systems must reason about causal relationships rather than timestamps.