Challenges
- Scale a web app to 10,000 RPSScale one web server and its database to carry a busy site.Chapter 1
- Estimate, then designTurn usage figures into a rate, then design for the peak.Method
- Search autocompleteSuggest the most popular searches after every character a person types.Challenge
- An order service that survives failuresKeep an order service serving when any one machine fails.Chapter 2
- Hotel bookingLet people browse rooms far more than they book, and never store a booking twice.Challenge
- A media portal for users worldwideServe readers on three continents quickly from one origin.Chapter 3
- Maps and routesServe map tiles, routes and drivers’ positions for millions of trips a day.Challenge
- Pastebin at scaleKeep millions of pastes, most of them read many times.Chapter 9
- Video sharingStream and take in videos, megabytes at a time, without the bytes swamping your servers.Challenge
- IoT telemetry ingestionStore a flood of device readings, nearly all of them writes.Chapter 5
- Unique IDsGive every record a unique ID that grows with time, faster than one database can hand them out.Challenge
- A sharded relational ledgerTake a ledger's writes past what one database leader can take.Chapter 4
- Digital walletMove money between wallets, recording every transfer exactly, at a rate no single leader can take.Challenge
- Rate limiterRefuse the calls over each client’s plan before they cost the servers anything.Challenge
- Social graph queriesAnswer friends-of-friends queries over a graph spread on many shards.Chapter 13
- Businesses nearbyFind the businesses near a person, from an index that is small and almost only read.Challenge
- Home timelines at scaleServe every user's home timeline, built as posts are written.Chapter 10
- Sales rank by categoryServe best-seller rankings when most shoppers want one category.Chapter 15
- A distributed cacheBuild a cache tier that survives losing a node.Chapter 14
- Key-value storeKeep shopping carts on many nodes, so a lost node stops nothing and every read sees the last change.Challenge
- Flash-sale catalog with a cold cacheKeep a flash sale's catalog up while its cache fills from cold.Chapter 6
- Call summaries for a contact centerSummarize every call as it ends, on your own GPUs, a hosted model, or both.Chapter 17
- A model too large for one GPURun an open model of 405 billion parameters on your own cloud account, to answer questions over long documents.Challenge
- Upload events with burstsTake bursts of upload events without dropping any.Chapter 7
- File syncKeep people’s files the same on all their devices, soon after every change.Challenge
- Nearby friendsShow people which friends are near, updated every few seconds, for everyone who has it on.Challenge
- ChatDeliver every chat message within a moment, to a million people online at once.Challenge
- Message queueCarry every service’s events to the consumers that handle them, each account’s in order.Challenge
- Stock exchangeMatch every symbol’s buy and sell orders in the order they arrived, without a pause.Challenge
- Ad click countsCount every click on every ad, and show advertisers their counts by the minute.Challenge
- Metrics and alertsKeep every server’s measurements, even while the store that holds them is down.Challenge
- NotificationsSend the company’s push notifications, emails and texts through providers that are sometimes slow.Challenge
- EmailSend mail to other providers, whose servers are sometimes slow or down, without losing any.Challenge
- Storing a web crawlKeep every unique page a crawler fetches, and no page twice.Chapter 11
- A ledger of bank transactionsRecord each bank transaction once, though some arrive twice.Chapter 12
- PaymentsCharge shoppers through an outside processor that is sometimes down, never charging twice.Challenge
- Checkout with a slow dependencyKeep browsing working while the service checkout calls slows down.Chapter 8
- Losing a zoneKeep a web app serving when a whole availability zone fails.Chapter 16
- Letter summaries on a bank’s own GPUsSummarize every customer letter on GPUs the bank runs itself, so that losing a zone loses no letter.Challenge
- Object storageBuild an object store for petabytes that almost never loses one, at the lowest price per gigabyte.Challenge