Plain Threads are the GOTO of todays computing
summary
Claiming plain threads are harmful, the keynote argues for futures/continuations and shows a working HPX runtime that builds execution trees instead of spawning OS threads, which lets small tasks scale: measured 25 million threads/sec with ~700 ns overhead per lightweight thread, and matrix-transpose benchmarks beat OpenMP on many-core and Xeon Phi setups. Code uses async, dataflow, futures and execution policies, the library is open source and the examples demonstrate concrete speedups and better resource utilization.
AI-written summary · about
chapters (11)
- 0:00The problem with threads
- 5:23Defining threads and concerns
- 10:41Why plain threads are harmful
- 15:23Performance limits and scaling
- 20:30The Four Horsemen of latency
- 29:32Introduction to HPX
- 34:11Futures and async programming
- 44:12Parallel algorithms in C++
- 49:30Resumable functions and await
- 56:26Distributed matrix example
- 1:12:13Conclusions and Q&A
related talks
01h 02mBack to Basics: Concurrency
23mIntroduction and pitfalls of Java's new concurrency mode
42mLet the Main Thread Breathe!
28mKeep the Main Thread Free With Web Workers
40mLudicrous Speed: Designing For Performance on the JVM
43mFrom CPU to GPU and FPGAs: Supercharging Java Applications with TornadoVM
32mSolid Ground
20mUsing Multithreading in the Browser