The one design in this series that scales in, not out. Why the fastest exchanges run on a single server, how an O(1) order book works, and why fairness is an engineering requirement — with an interactive matching engine.

Read the full article →

Four designs, each one fixing what the last one broke. In-memory sharding, two-phase commit, TC/C and Saga, then event sourcing — and why the only data that truly needs to be durable is the event log. With an interactive replay you can scrub.

Read the full article →

Ten transactions per second, and the hardest correctness problem in the series. Double-entry bookkeeping, exactly-once as retry plus idempotency, and why reconciliation is the last line of defence — with an interactive ledger you can't unbalance.

Read the full article →

Ranking 25 million players in real time. Why SQL cannot do it, how a skip list makes it O(log n), and the sharding problem that has no clean answer — with a live leaderboard you can play with.

Read the full article →

Eleven nines of durability on drives that fail constantly. Block versus file versus object storage, why small files waste disks, and the erasure-coding trade that is not as simple as it looks — with an interactive durability calculator.

Read the full article →

A billion users, two exabytes a year, and protocols designed in the 1980s. Why email storage dwarfs everything else in this series, how denormalisation answers a query NoSQL refuses, and the part that isn't engineering at all — with an interactive deliverability checker.

Read the full article →

Three reservations per second — and it is the hardest design in the series. Double booking, pessimistic versus optimistic locking, and the one-line fix the textbook leaves out. With an interactive race-condition simulator you can try to break.

Read the full article →

A billion clicks a day, and the numbers become invoices. Event time versus processing time, watermarks, exactly-once as an actual requirement rather than a nice-to-have — with an interactive watermark simulator for the accuracy-versus-latency trade at the heart of every streaming system.

Read the full article →

Ten million metrics, one year of retention, and a write load that never stops. Why a general-purpose database is the wrong answer, how Gorilla gets 16 bytes down to 1.37, and the cardinality bomb that kills more monitoring systems than any other cause — with an interactive calculator.

Read the full article →

Why a plain append-only file beats a database, how partitions turn ordering into a tunable knob, and what exactly-once really costs. Includes an interactive delivery-semantics explorer — and the fourth quadrant nobody mentions, where you get neither guarantee.

Read the full article →

One billion daily users, 100 petabytes of map tiles, and a road graph too big to fit in memory. Routing tiles, the CDN economics of map rendering, adaptive rerouting — plus an interactive tile explorer, and why the real tile URL is not what the textbook says.

Read the full article →

333,000 location updates a second turn into 13.3 million messages a second. Why moving people are a completely different problem from static businesses — WebSocket fan-out, Redis Pub/Sub as a routing layer, and an interactive calculator for the amplification that decides the whole design.

Read the full article →

How does 'restaurants near me' return results in under 100ms from 200 million businesses? Geohash, quadtrees, Google S2 and Uber's H3 explained from first principles — with an interactive geohash encoder you can play with, and the boundary bug that breaks the obvious implementation.

Read the full article →

File sync for 10 million users and 500 petabytes. Delta sync, why fixed-size blocks break the moment you insert a byte, content-defined chunking, conflict resolution that never loses data, and the deduplication trick that leaks information.

Read the full article →

A video platform where the CDN bill dwarfs everything else. Transcoding as a DAG, adaptive bitrate ladders, chunked resumable uploads, and the cost optimisations — per-title encoding, AV1, ISP peering — that decide whether the business works.

Read the full article →

Suggestions in under 100ms, on every keystroke, for 10 million users. Tries with cached top-k, why the ranking happens offline, the trending-query problem the weekly rebuild cannot solve, and the client-side trick that removes three quarters of your traffic.

Read the full article →

Build a Messenger-style chat system for 50 million daily users. Why WebSocket beats polling, how server-assigned sequence numbers guarantee ordering, multi-device sync, presence with heartbeats, and the reconnect problems that only appear in production.

Read the full article →

Fan-out on write or fan-out on read? Build a Facebook-style feed for 10 million daily users, understand exactly why one celebrity account breaks the obvious design, and see the hybrid that every large platform actually ships.

Read the full article →

Push, SMS and email at 16 million messages a day. Per-channel queues, why exactly-once delivery is impossible and what to do instead, device-token hygiene, priority lanes that keep OTPs away from marketing blasts, and the retry logic the textbooks skip.

Read the full article →

Crawl a billion pages a month without getting banned. The URL frontier that enforces politeness and priority, BFS over the web graph, spider traps, near-duplicate detection with SimHash, and what RFC 9309 changed about robots.txt.

Read the full article →

Design a TinyURL-style service handling 100 million new links a day. Base62 encoding, why 7 characters is exactly right, the 301-versus-302 decision that quietly breaks your analytics, and the abuse problem the textbooks skip.

Read the full article →

How to generate billions of unique, sortable, 64-bit IDs per day with no central coordination. Multi-master replication, UUIDs, ticket servers and Twitter Snowflake explained from first principles — plus UUIDv7, ULID and how to choose between them in 2026.

Read the full article →

A complete beginner-friendly guide to designing a distributed key-value store — covering CAP theorem, quorum consensus, vector clocks, gossip protocol, Merkle trees, and the write/read paths inspired by DynamoDB and Cassandra.

Read the full article →

A complete, beginner-friendly deep dive into designing a rate limiter — covering all 5 algorithms, Redis architecture, distributed challenges, and real patterns from AWS, Stripe, Shopify & Cloudflare.

Read the full article →