Techniques and mechanisms that coordinate simultaneous access to shared resources so that operations preserve correctness, isolation, and intended invariants in the presence of overlapping execution.

Principle

Enforce a serialization or equivalent ordering of conflicting operations (via locks, timestamps, versioning, optimistic validation, or transactional protocols) so that interleaved executions are observationally equivalent to some correct serial execution or to a specified isolation level.

Demonstration

A database system uses two-phase locking: Transaction A acquires a write lock on account X before debit, Transaction B blocks on that lock until A commits, preventing lost-update anomalies and ensuring account balance invariant.

Misapplication

Applying coarse-grained mutual exclusion across unrelated resources to avoid race conditions; this prevents races but unnecessarily serializes independent operations, causing throughput collapse and unfair latency.

Consequence

When correctly applied, shared-state anomalies (lost updates, dirty reads, non-repeatable reads) are prevented and program invariants are preserved, at the cost of increased coordination overhead and potential contention or deadlocks if not managed.

Reversal

No concurrency control: concurrent operations proceed without coordination, producing races, non-deterministic outcomes, lost updates, and invariant violations that depend on scheduling.

Boundary

Scope includes multi-threaded programs, database transactions, and distributed stores where mutable shared state is accessed concurrently. Excludes purely immutable data access and entirely sequential single-threaded execution. Guarantees depend on chosen isolation semantics and system model (synchronous vs asynchronous, fail-stop vs Byzantine).

Semantic Tension

Tension between strict serializability (maximal correctness) and throughput/latency: pessimistic locking gives strong isolation but high contention; optimistic methods improve throughput but may increase aborts. Also trade-off between local concurrency control and global coordination in distributed environments.

Synthesis

Concurrency control comprises mechanisms that transform overlapping access patterns into executions that satisfy chosen correctness criteria by detecting and ordering conflicting operations, balancing protection of invariants against performance costs.