Optimistic concurrency control
Optimistic concurrency control executes transaction work speculatively and validates that it can commit under the required isolation guarantees, aborting attempts that fail validation.
Mechanism: The classic design separates reading and private updates, validation, and publication of accepted writes. Its serializability checks consider read sets, write sets and overlapping transaction phases, not only competing writes to one object. [1]
Isolation: Optimistic checking and multiversion storage are separate choices. Larson et al. describe both optimistic and pessimistic multiversion methods. Their serializable optimistic method validates read versions and scans for phantoms; its snapshot-isolation mode omits those read and scan checks. These are rules of that design, not every engine. [2] SQL Server memory-optimized tables can still perform read and phantom validation under Snapshot isolation to enforce unique and foreign-key constraints. [3]
Application: An atomic row-version-checked update guards that row, not every dependency used to decide its new value. The required validation scope follows the transaction's isolation contract. Successful validation also does not by itself establish durable storage. [2] [3]
Terminology: Optimistic crash consistency concerns persistence ordering and recovery, not this transaction concurrency-control policy. [4]
Relations
- related
- Hyder, Serializability, Snapshot isolation, Write skew, Silo, Tango
Limitations
- Retaining validation history has a storage cost; discarding required history can force a restart even without a demonstrated conflict. Repeated restarts can cause starvation. [1]
Citations
- Kung and Robinson. On Optimistic Methods for Concurrency Control. ACM TODS 6(2), 213-226 (1981), sections 2-5. DOI: 10.1145/319566.319567.
- Larson et al. High-Performance Concurrency Control Mechanisms for Main-Memory Databases. PVLDB 5(4), 298-309 (2011), sections 3 and 6.
- Microsoft: Transactions with memory-optimized tables, validation, commit processing, conflict detection and retry logic.
- Chidambaram et al. Optimistic Crash Consistency. SOSP 2013, sections 4-5. DOI: 10.1145/2517349.2522726.
Further reading
Sources & caveats
- Primary-source synthesis; no native engine execution, performance comparison or crash experiment.
- The 1981 paper gives serializability algorithms, not a universal lock-free or bounded-retry guarantee. Later implementations have different isolation and commit contracts.