The delaying the future approach
summary
Defines a concrete Delaying the Future theorem that lets a run be transformed by postponing every event outside the past of a chosen cut by delta while remaining indistinguishable to processes; proves this for asynchronous message passing (uses happens‑before/message‑chain) and for TSO by defining a new occurs-before relation. Consequences: non‑commuting operations must be connected by those relations, every completing register op needs message chains to at least f other processes under f crashes, and under TSO a writer must use a fence or RMW (or observe a feedback loop) to ensure its write is visible.
AI-written summary · about
chapters (17)
- 0:00Delaying the future approach
- 1:24Operation order in async models
- 3:12Real-time order challenges
- 5:01Future theorem explained
- 6:26Future deal in message passing
- 13:11TSO weak memory model
- 17:15Occurs-before in TSO
- 21:09Impact on implementations
- 21:55Redefining commutativity
- 24:25Linearizable registers in TSO
- 26:53Linearizability lower bounds
- 29:00Future deal implications
- 30:45Conclusion
- 32:51Q&A on TSO
- 34:18Extending to other models
- 36:35Comparison with happens-before
- 38:43Final Q&A
related talks
36mThe FIDS Theorems: Tensions between Multinode and Multicore Performance in Transactional Systems
21mEnabling User-level Asynchronous Tasking in the FA-BSP Model - Case Study:(…)
38mEventual Consistency: A Question of Time and Expectations
42mTime-sensitive protocol: Links From Design to Implementation
02h 42m!!! Do not publish or share !!!!
27mContinuous Tasks and the Asynchronous Computability Theorem
35mCommunicating Sequential Processes
14mA Practical Approach to Multi-Concurrent Proposers