The goal is to exploit parallelism to maximize TPS (transactions per second).

Anomalies

Lost update

Two transactions read the same data, but one update gets overwritten and lost.

Sequence: R1 → R2 → W1 → W2

• T1: Reads X = 100
• T2: Reads X = 100
• T2: Updates X = 100 – 20 = 80 → Writes X = 80
• T1: Updates X = 100 + 50 = 150 → Writes X = 150

Dirty read

A transaction reads uncommitted changes of another transaction.

Sequence: R1 → W1 → R2 → A1 → W2

• T1: Reads X = 100
• T1: Updates X = 200 (but doesn’t write)
• T2: Reads X = 200 (thinks balance is 200)
• T1: Aborts → Rollback → X = 100

Non-repeatable reads

A transaction reads the same row twice but gets different values because another transaction updated it in between.

Sequence: R1 → R2 → W2 → R1

• T1: Reads X = 100