Course

System Design Lab

The substrate under the designs — data, distribution, and the tradeoffs you get grilled on. Companion to the chapter decks.

0%

0 of 36 written lessons complete

Start where you left off

Drills → · cards due · spaced repetition over every lesson Glossary → 117 terms, searchable with Ctrl+K

0 Foundations

The vocabulary. Brisk — you know most of this already.

Latency Numbers & the Estimation Reflex Turning a vague product description into servers, gigabytes, and gigabits 18 min Networking for System Design DNS, handshakes, HTTP versions, and why the first byte takes so long 17 min Load Balancing L4 vs L7, the algorithms that matter, and how health checks cause outages 16 min Caching Where to put it, how to invalidate it, and the four ways it takes you down 20 min CDN & the Edge Moving bytes closer to users, and the cache key that decides whether it works 15 min Availability Math Nines, serial dependencies, and why redundancy usually disappoints 15 min

1 Data

Storage engines, transactions, replication, partitioning.

Storage Engines — B-trees vs LSM-trees Why two databases with the same API differ by 100× on writes 19 min Indexing Why you added an index and the query got no faster 18 min Transactions & Isolation Levels Dirty reads, write skew, and the anomaly that snapshot isolation won't save you from 20 min Replication One leader, many leaders, or none — and what each one costs you 18 min Partitioning & Consistent Hashing Splitting data across machines without reshuffling the world every time you add one 20 min Choosing a Datastore Access patterns first, database second — and why the answer is usually Postgres 18 min

2 Distributed Core

Clocks, CAP, consistency, consensus. The advanced spine.

Failure, Timeouts & Partial Failure Why you can never tell a dead server from a slow one, and what follows from that 18 min Clocks & Ordering Ordering events without a global clock — Lamport, vectors, HLC, TrueTime 18 min The Consistency Spectrum Linearizable, sequential, causal, eventual — and which one you can actually afford 18 min CAP, Properly — and PACELC Why "we chose AP" is not an answer, and what the theorem actually forbids 17 min Quorums, Repair & Anti-Entropy R + W > N, what it actually buys, and the three ways it quietly stops being true 19 min Consensus & Raft How a cluster agrees on one ordered log, and why you should use as little of it as possible 21 min Locks, Leases & Leader Election Why your distributed lock doesn't lock, and the one mechanism that fixes it 17 min Idempotency & the Exactly-Once Myth Exactly-once delivery is impossible; exactly-once effect is a design choice 18 min

3 Patterns & Assembly

Queues, events, resilience, multi-region, observability.

Queues, Logs & Backpressure A queue forgets on read, a log forgets on a clock — everything else follows from that 19 min Event-Driven Architecture Events, commands, CQRS, event sourcing and sagas — and which of them you actually need 19 min API Design at Scale Pagination, versions and error shape are where public APIs quietly fall over 18 min Resilience Patterns Breakers, bulkheads, shedding and degradation — containing a failure already underway 18 min Search & Inverted Indexes Why the index is inverted, why BM25 replaced TF-IDF, and why page 1000 is expensive 18 min Stream Processing Event time, watermarks, and deciding when an unbounded computation is allowed to answer 18 min Multi-Region & Cell-Based Architecture Three different reasons to leave one region, and why the modern answer is cells rather than continents 19 min Observability & SLOs Three signals, burn-rate alerting, and why the dashboards stay green during the outage 18 min Multi-Tenancy Isolation models, the noisy neighbour, and the operations nobody plans for 18 min

4 Case Studies

What the chapter decks do not already cover.

Case Study — Discord's Message Store How a two-field partition key holds trillions of messages, and why the database underneath it was replaced 16 min Case Study — Netflix Open Connect A CDN you give away, a cache that is never cold, and failure injected on purpose 16 min Case Study — Stripe's API Idempotency keys, rolling versions, and what it costs to promise you will never break 16 min Case Study — Dynamo & DynamoDB How one shopping-cart requirement produced a technique catalogue, and one idea that did not survive 17 min Case Study — Google Spanner Buying global strict serializability with atomic clocks and deliberate waiting 17 min

5 Interview Track

The framework, timed mocks, rubrics, tradeoff drills.

The 45-Minute Framework How to spend the clock, what to say when, and how to drive the conversation 19 min How Candidates Actually Lose Twelve failure modes, in rough order of how often they decide the outcome 16 min