Why Distributed Systems Are Hard: Space, Time, and Consensus Explained

Distributed systems arise when multiple machines must cooperate to complete a single task, and that requirement strips away the shared memory, unified clock, and single lock that a solo machine provides for free. The core difficulties trace back to three root problems: space, meaning no single node holds the complete picture; time, meaning there is no global clock and networks reorder messages unpredictably; and consensus, which emerges when the first two problems collide. Replication strategies such as single-leader, multi-leader, and leaderless designs each resolve the write-conflict problem differently but introduce their own failure modes, while partitioning by key range or hash creates further trade-offs between write balance and query efficiency. Because there is no reliable wall clock across machines, engineers use logical and vector clocks to track causality rather than absolute time, though vector clocks grow costly at scale. These compounding challenges explain why distributed systems offer a spectrum of consistency models — from strict linearizability to eventual consistency — each trading correctness guarantees for performance or availability.