Performance Architecture
summary
Describes converting a 5+-year .NET codebase to reach orders‑of‑magnitude speedups by changing data layout and memory strategy: replacing JSON parsing with a blittable format to avoid repeated allocations, using memory‑mapped files and an arena allocator to reuse buffers, optimizing routing and string escaping hot paths, and trading some backward compatibility for performance. Benchmarks show millions of requests in seconds and sustained ~50,000 requests/sec on a small Graviton node, and the rebuilt system behaves consistently under load.
AI-written summary · about
chapters (15)
- 0:00Start
- 2:57Timeline of Our Releases
- 6:18Memory and Gc Pressure
- 6:43Second System Syndrome
- 8:02How Can I Be Fast
- 23:59Micro Optimizations
- 26:02How Do We Change the Actual Architecture
- 29:14Pointer Arithmetic
- 35:38How the Gc Works
- 36:48Arena Allocator
- 37:49Live Actions
- 41:32Page Cache
- 42:49Escape of Strings
- 43:52Questions
- 47:11What's Your Opinion about Making Garbage Collection in Dot Net Optional and Using Reference Counting or Resource Acquisition
related talks
01h 00m.NET Rocks Live – Modernizing .NET Applications
38mAccelerate
53mPowering 100+ million daily users
38mPerformance in the Wild
27mEngineering Productivity Starts with Predictable Behaviour
30mArchitecture & Teams at Less-than-Google Scale
42mPanel - Fortune 100 Graphs - Powering Today's Need and Tomorrow's Trends
48mRewriting Parse.com for scale