# Concurrency

lock contention, thread/worker sizing, race conditions.

## Measure

- concurrency profiling → locks, critical sections
- thread dump analysis → states, blocked, deadlocks
- lock analysis → hold times, wait times, contention rate
- metrics → lock wait, thread utilization, context switch rate, contention ratio

## Optimize

- lock-free data structures → atomics over locks
- reduce critical sections → minimize locked work
- tune worker pools → match pool to workload (CPU vs I/O bound)
- async processing → offload blocking work
- partition data → sharding reduces shared structure contention
- read-write locks → read-heavy workloads

## Pitfalls

- premature lock removal → complex, error-prone; only when contention measured
- over-parallelism → context switch overhead
- false sharing → adjacent writes invalidate cache lines
- deadlock risk → changing lock order; test thoroughly