Byte-Sized Design

Byte-Sized Design is a Substack that provides engineering summaries from major tech companies, covering innovations and strategies in system design, security measures, authentication technologies, database management, coding practices, API development, and testing methodologies. It includes case studies from companies like Tinder, Uber, Airbnb, Yelp, Google, and Netflix.

System Design Security and Safety Authentication Technologies Database Management Coding Practices API Development Testing Methodologies Software Architecture Performance Optimization Scalability Solutions

The hottest Substack posts of Byte-Sized Design

And their main takeaways

The Byte-sized Design List of System Design Blogs

491 implied HN points • 15 Jan 24
  1. The post lists many system design blogs for readers to follow.
  2. The Byte-Sized Design newsletter is looking for writers to help maintain their content.
  3. There's a 7-day free trial available for readers who want full access to the post archives.

The Database Google Uses to Build for Everyone

157 implied HN points • 08 Feb 24
  1. Google built a distributed database called Spanner to handle demanding uptime and scale requirements globally.
  2. Spanner achieves externally consistent reads by using snapshot-based reads and a distributed clock called TrueTime.
  3. Spanner assigns timestamps to transactions and ensures order consistency, overcoming clock skew issues in distributed systems.

Uber's Emergency Button

176 implied HN points • 21 Jan 24
  1. Uber introduced an emergency button for riders and drivers to contact emergency dispatchers discreetly
  2. The emergency button integration involves sharing location details immediately with RapidSOS and using real-time location services like Kafka
  3. Safety and security are top priorities for Uber to protect users during ride shares

Why AirBnB Won't Charge You Twice

157 implied HN points • 28 Jan 24
  1. AirBnB uses idempotency to prevent double charges
  2. Give every request a unique idempotent key to avoid duplicates
  3. Clean up the database periodically to manage requests

Passkeys Your Company is Making You Use.

176 implied HN points • 08 Jan 24
  1. Passwords are unsecure because people tend to reuse them.
  2. Using passkeys for authentication is usually more secure than using just passwords.
  3. Passkeys work by using public key cryptography with a private and a public key.
Get a weekly roundup of the best Substack posts, by hacker news affinity:

Yelp knows their resources are better than your phone.

157 implied HN points • 25 Dec 23
  1. Yelp optimized their server-side rendering by using worker threads instead of the main thread.
  2. Yelp switches to client-side rendering if server-side rendering fails to improve efficiency.
  3. Yelp's improvements led to reduced rendering time, faster service startup, enhanced observability, and a more extensible system.

Uploading Instagram Videos with 94% LESS CPU

58 implied HN points • 11 Feb 24
🕹 Technology Social media Video Compression Engineering Digital
  1. Instagram improved video uploading efficiency by compressing videos first to progressive encodings and then converting them to adaptive-bit-rate videos, saving 94% of resources.
  2. The challenge for Instagram was to support various video formats for different devices while minimizing resource consumption and CPU usage.
  3. Insightful optimization was achieved by realizing that progressive and adaptive bit rate encodings could use the same codec, streamlining the encoding process to focus on scalability.

Breaking the limits of 10 million concurrent Discord users.

137 implied HN points • 11 Dec 23
  1. Discord optimizes by sending events only to active users on the server.
  2. Discord splits work from a single process to multiple sub-processes for efficiency.
  3. Discord uses in-memory databases to store member data and improve performance.

GitHub Copilot Uses Your Code for Their APIs

137 implied HN points • 03 Dec 23
  1. GitHub Copilot suggests code based on your open tabs, but can only read 6,000 characters at a time.
  2. Algorithms prioritize code snippets and comments to create prompts for Copilot suggestions.
  3. Copilot uses Vector Databases to speed up API calls, making developers 55% faster.

Yelp Abuses Binary Search for Merging Code

157 implied HN points • 10 Nov 23
  1. Yelp used binary search to find issues in merging code efficiently.
  2. Yelp's decision to have a Monorepo was driven by the need to update all code at once for easier maintenance.
  3. Yelp's pipeline involved grouping and merging code changes in batches to improve efficiency.

Passwords are dead. Shopify wants Passkeys

137 implied HN points • 29 Oct 23
  1. Shopify wants to replace email and SMS verification.
  2. Passkeys use devices like pins or thumbprints for authentication.
  3. You can authenticate with multiple devices or register new passkeys.

How Message on Slack Really Works

137 implied HN points • 09 Oct 23
  1. Slack sends tons of messages daily using a server setup with channel servers.
  2. Slack's architecture involves user data fetching, web socket connection, and event processing.
  3. The feature of knowing when someone is typing on Slack is handled as an 'event' similar to how messages are processed.

Matching Tinder API Services ❤️

216 implied HN points • 23 Jul 23
  1. Tinder built its own API Gateway to consolidate services and improve scalability.
  2. An API Gateway acts as an intermediary server, handling requests before passing them to actual servers.
  3. Building a customized API Gateway like Tinder's can enhance development velocity and meet specific requirements.

The Disney architecture for India Vs. Pakistan World Cup

98 implied HN points • 16 Oct 23
  1. Disney faced challenges with storing user viewing data in their architecture
  2. They scaled their infrastructure by migrating to ScyllaDB for faster reads and writes
  3. The migration strategy involved writing to both databases, then switching all operations to the new one

LinkedIn Literally Stops Spamming APIs

98 implied HN points • 23 Sep 23
  1. LinkedIn stopped creating separate APIs for each company to onboard
  2. They switched to using GraphQL to handle varying API requests from different companies
  3. Implementing GraphQL reduced API development time from 2 quarters to 2 days

Netflix is Chill with Breaking Their Services

137 implied HN points • 15 May 23
  1. Netflix uses a middle tier server called MAP to manage and simplify their multiple services
  2. MAP allows Netflix to centralize logic and avoid complicated client-side processing
  3. Netflix conducts thorough testing, including intentional failure testing, to ensure MAP functions smoothly

Why Reusing Code Backfired Uploading Photos on Dropbox

117 implied HN points • 23 Apr 23
  1. Reusing the same code for Android and IOS can lead to issues with different device requirements.
  2. Dropbox had to switch to native platform technologies like Swift and Kotlin for uploading photos.
  3. To improve, they focused on background activity, speed, and validating their new implementations.

Lyft Tests In Production 🚗💨

78 implied HN points • 10 Jul 23
  1. Lyft tests their services in production to get more realistic results and save costs.
  2. Testing in staging environments can be expensive and risky for companies like Lyft due to the inability to simulate real-world traffic.
  3. Lyft simulates traffic in production using a configuration that defines random events for users, making it a safer and more efficient testing method.

Google Went Monorepo And Never Looked Back

98 implied HN points • 28 May 23
  1. Google uses a monorepo for all their code, making it simpler with less dependency management.
  2. Google employs trunk-based development and a strict coding workflow to manage changes effectively.
  3. Advantages of Google's monorepo include a single source of truth, better dependency management, and atomic changes, but it also has challenges like handling random dependencies and requiring specific tools.

Over 90% Storage Reduction at Quora

39 implied HN points • 09 Sep 23
  1. Quora implemented simple caching strategies for over 90% storage reduction.
  2. Optimizing cache keys and implementing range based caching resulted in big performance gains.
  3. Queries for specific user languages and question redirections were optimized for efficiency.

How Discord Stores 1 Trillion Messages (12 Zeroes!)

78 implied HN points • 19 Mar 23
  1. Discord migrated from Cassandra to ScyllaDB to handle trillion messages more efficiently.
  2. The move was necessary due to latency issues and concurrent scalability challenges.
  3. ScyllaDB's advantages included no garbage collector and support for reverse querying.

Ebay doesn't use E2E testing for their APIs...

39 implied HN points • 19 Jun 23
  1. Ebay uses contract testing instead of E2E testing for their APIs
  2. Contract testing involves consumers providing API tests for updates
  3. Tools like Pactflow simplify the contract testing process

Levels.fyi Used WHAT for their backend?

3 HN points • 04 Feb 24
  1. Levels.fyi initially used Google Sheets as their backend
  2. Challenges with using Google Sheets included slower page load time and API limits
  3. To overcome limitations, they duplicated write calls to new database server and slowly migrated read API calls

DoorDash Breaks Their Own Services?

19 implied HN points • 12 Jun 23
  1. DoorDash tests their microservices by intentionally breaking their own services in local and staging environments.
  2. DoorDash created a fault injection tool called Filibuster to test all possible error scenarios with their services.
  3. Filibuster ensures automated testing of permutations of failed dependencies, providing developers insights on potential errors to handle.

Twitter Made Searching Scalable Before Elon Musk

19 implied HN points • 02 Apr 23
  1. Twitter faced challenges with using Elasticsearch for search and backfill operations.
  2. They addressed the issue by offloading traffic and backfill requests to separate services.
  3. The new services included a proxy for routing requests, a backfill service, and an ingestion service.