Enabling User-level Asynchronous Tasking in the FA-BSP Model - Case Study:(…)
summary
Extends FA-BSP into a fine-grained FA-BSP (FABS) runtime that adds actor-based asynchronous messaging and tests enabling user-level async+finish tasking on distributed triangle counting. Uses a Convey-based runtime, compares single-worker vs two-worker handlers, and contrasts per-message blocking finish with non-blocking task creation plus a final global finish. Chunked messages reduce overhead; two-worker chunked runs outperform blocking and single-worker versions. Reported speedups range 3%–32% on an HPC cluster but require careful hand tuning.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:47Classic BSP model limits
- 2:33FA-BSP with async messaging
- 4:00FA-BSP execution example
- 5:25Performance results
- 7:12Extending FA-BSP to threading
- 9:19Blocking vs non-blocking
- 12:12Runtime challenges
- 13:37Triangle counting setup
- 15:26Results analysis
- 17:32Conclusion
- 18:36Acknowledgements
related talks
32mMaximizing Throughput and Scalability for Akka Streams
22mAsynchronous Computing at Scale: Overview and Learning
01h 07mThe Asynchronous C++ Parallel Programming Model
17mTowards High-Performance Transactional Stateful Serverless Workflows with...
15mAlea-BFT: Practical Asynchronous Byzantine Fault Tolerance
14mFeedback-guided Adaptive Testing of Distributed Systems Designs
23mA Haskell Auto-Parallelizer for Distributed Computing
01h 02mA Unifying Abstraction for Async in C++