Distributed System Illustrated

About This Book

Distributed System Illustrated: A Visual Guide for Engineers

Building reliable systems out of unreliable components — that's the essential challenge of distributed computing. Networks drop packets, clocks drift, nodes crash, and yet we expect 24/7 availability from every modern service.

This book takes a visual, first-principles approach to distributed systems. Instead of drowning you in formal proofs or assuming you already understand the jargon, it uses clear diagrams and intuitive reasoning to peel back the layers — from why physical clocks can't be trusted, all the way to how Google Spanner achieves global consistency with atomic clocks.

• Derive, don't memorize. The Paxos algorithm isn't presented as a finished artifact to be accepted on faith. Instead, you'll watch it emerge step by step from simple quorum reads and writes, understanding why each piece is necessary before moving to the next.
• Engineering depth, not just theory. Raft is covered in full — not just leader election and log replication, but also cluster membership changes (single-step and joint consensus), liveness problems under network partitioning (Pre-Vote, CheckQuorum, Leader Lease), log compaction, and implementing linearizable reads.
• Visual-first. Every concept is illustrated. Consistency models, log replication flows, two-phase commit sequences, TrueTime commit wait — all accompanied by diagrams that make the abstract concrete.
• Full-stack coverage. From the physics of quartz clocks and NTP's flawed symmetry assumption, through Lamport clocks and vector clocks, replication and CAP, Paxos and Raft, consistent hashing and Dynamo, all the way to distributed transactions, Spanner's TrueTime, TCC, and SAGA.

• Why physical clocks fail in distributed systems, and how logical clocks (Lamport, Vector) capture causality instead
• The complete consistency model spectrum — from linearizability to eventual consistency — and when to choose which
• How Paxos works, derived from first principles via quorum read-write evolution
• Raft in production-grade detail: election, replication, safety, membership changes, liveness, snapshots, and linearizable reads
• Consistent hashing, virtual nodes, and how Dynamo handles partitioning at scale
• ACID internals (Undo/Redo logs, 2PL, OCC, MVCC, Write Skew), distributed transactions (2PC/3PC), and how Spanner combines TrueTime + Paxos + 2PC for global external consistency
• When to use strong consistency (2PC/Spanner) vs. eventual consistency (TCC/SAGA), and the trade-offs involved

• Backend engineers and architects who want to go beyond surface-level understanding of distributed systems
• Developers preparing for system design interviews who need solid command of Paxos, Raft, CAP, and consistency models
• Students looking for a reference that's more visual than a textbook and more rigorous than a blog post
• Anyone curious about how large-scale internet systems actually work under the hood

Basic computer science knowledge (data structures, operating systems, networking) is assumed. No prior distributed systems experience required.

1. Overview of Distributed Systems — Challenges, fallacies, and the mindset shift from single-machine to distributed thinking
2. Distributed System Models — Two Generals, Byzantine Generals, communication/failure/timing models
3. Time and Order — Physical clock limitations, Happens-Before, partial vs. total order, Lamport clocks, vector clocks
4. Replication — Replication modes, quorum read-write, consistency models (linearizability to eventual), CAP and PACELC
5. Distributed Consensus Algorithms — FLP impossibility, Paxos derived from first principles, Raft in full engineering detail
6. Partitioning — Consistent hashing, Chord, virtual nodes, Amazon Dynamo
7. Transactions — ACID internals, concurrency control (2PL, OCC, MVCC), distributed transactions (2PC/3PC), Google Spanner, TCC, SAGA