Performance trade-offs in distributed replication and consensus — write latency, tail latency, write vs read scalability, consensus throughput limits, batching and pipelining, geographic distribution costs, and backpressure design.
Paxos vs Raft explained — how both consensus algorithms work, why Paxos is hard to implement, how Raft’s leader election and log replication work step by step, and why Raft dominates production systems like etcd, CockroachDB, and TiKV.
Why is consensus hard in distributed systems? Learn the Two Generals Problem, FLP impossibility, crash-stop vs Byzantine failures, and why consensus must stay off the data path — with production examples from etcd, Raft, and Google Spanner.