Fast Atomicity Monitoring
summary
Presents AtomSanitizer, an algorithm for conflict serializability that reduces worst-case time to O(nk^2) by tracking approximate latest-remote and earliest-local transactions via two vector clocks, detecting cycles with three concrete conditions. Implements both offline Java tracing and an integrated into ThreadSanitizer online monitor, processing traces up to ~50 GB, beating RegionTrack/Aerodrome in runtime and memory, and reporting only 1.3× runtime overhead over TSAN while remaining the first practical AtomSanitizer for real-time atomicity monitoring.
AI-written summary · about
chapters (9)
- 0:00Atomicity monitoring intro
- 1:37Static analysis limits
- 2:21Dynamic concurrency analysis
- 4:27Conflict serializability
- 6:15Voldemort algorithm inefficiency
- 7:40New fast algorithm
- 9:12Key lemma
- 11:20Experimental results
- 12:46Question and answer session
related talks
20mAtomiS: Data-Centric Synchronization Made Practical
19mRuntime Detection of Data Races in OCaml with ThreadSanitizer
36mATOMICALLY { DELETE YOUR ACTORS }
48mThreadSanitizer and Heap Sampling
38mKernel Sanitizers Office Hours
01h 04mDebugging data races
52mNonblocking algorithms/ CAS/Atomics
01h 22mBeyond Sequential Consistency - Leveraging Atomics for Fun & Profit