Developing the Bloomberg Terminal -- Local performance & measurement techniques

on YouTube ↗01h 04m

" summary

Bloomberg Terminal combines a C++ data layer with embedded Chromium/V8 to handle ~10–50k updates/sec across 100–200 windows; the pipeline uses a compact NaN-boxing datum variant, polymorphic allocators and a monotonic buffer resource to cut heap calls (example: an 80-char unique‑chars test drops from 86 mallocs and 13.3μs to one allocation then to zero by using a stack buffer). Tooling records and replays live push traffic for reproducing issues; WebAssembly prototypes run within ~40% of native and sometimes beat the native/JS boundary.

AI-written summary · about

" chapters (51)

1. 0:00Introduction
2. 1:56Agenda
3. 3:09Background
4. 4:40What is it
5. 5:41Server side
6. 7:36Backend
7. 8:26Lua scripting engine
8. 9:05Application container processes
9. 9:40Metrics subsystem
10. 11:06Data layer
11. 11:51Data layer components
12. 12:53Combining data
13. 13:41DataNow Inspector
14. 14:21Datum
15. 15:08Custom data types
16. 16:21What is datum
17. 17:08Can we do better
18. 17:50Floatingpoint format
19. 18:55String optimization
20. 20:18Polymorphic allocators
21. 23:15How does this work
22. 24:48Custom allocator
23. 26:41Monotonic Buffer
24. 31:14Data Layer Performance
25. 31:57Portable Columns
26. 32:48Worksheet Expressions
27. 33:54Coalescing
28. 35:00Chromium
29. 35:19UI
30. 38:04Bindings
31. 38:47Challenges
32. 40:33Minimize GC
33. 42:02Life cycle
34. 42:33Application
35. 43:04Disposal pattern
36. 46:23Overall CPU load
37. 46:40Bloomberg app
38. 47:22Bloomberg 3D view
39. 48:01Memory usage
40. 49:10Cake view
41. 49:40Live memory usage
42. 50:06Measuring responsiveness
43. 51:22Instant message system
44. 51:36JavaScript levels
45. 52:59UI Timing
46. 56:38Playable
47. 57:00Free integration testing
48. 57:29WebAssembly
49. 58:30What is WebAssembly
50. 59:13C toolchain
51. 59:46How do we hook up to it

" related talks

• 53mBloomberg Becomes BrowserJason Williams, Paul Williams · performance.now() 2024
• 47mBenchmarking C++, From video games to algorithmic tradingAlexander Radchenko · Meeting C++ 2018
• 33mReaching Ludicrous SpeedMatteo Collina · Codemotion Amsterdam 2016
• 01h 29mWhen Nanoseconds Matter: Ultrafast Trading Systems in C++David Gross · CppCon 2024
• 30mPerformance Tuning SecretsPeter Flynn · JSConf US 2013
• 42mSelf-Awareness in C++ CodePaul Williams · NDC TechTown 2018
• 39mMicro BenchmarksStoyan Stefanov · performance.now() 2022
• 17mVisualising Front-End Performance BottlenecksRichie McColl · React Summit 2020