PACELC Theorem

The PACELC theorem extends the CAP theorem by describing the trade-off a distributed system makes even when the network is healthy. It reads: if there is a Partition (P), choose between Availability (A) and Consistency (C); Else (E), when the system runs normally, choose between Latency (L) and Consistency (C).

PACELC splits reasoning into two situations:

• During a partition (PA vs PC): identical to CAP's availability-versus-consistency choice.
• Else, no partition (EL vs EC): to guarantee that every replica agrees on a value, a write must reach and be acknowledged by multiple replicas before returning. That coordination adds latency. A system can instead return quickly after one replica accepts the write (low latency) and propagate the change asynchronously, which weakens consistency.

Systems are classified by both halves, such as PA/EL or PC/EC. The Else branch is really a statement about replication: waiting for more replicas raises both durability and consistency but costs round trips, while replying early trades that guarantee for speed.

The Else branch is a dial, not a switch. Requiring acknowledgment from a quorum of replicas moves you toward EC and adds the latency of the slowest replica in the quorum; acknowledging after a single replica moves you toward EL and risks serving values that have not yet propagated. The same store can sit at different points per operation, letting a critical write pay for consistency while a cheap read stays fast.

• Introduce PACELC right after CAP to cover the normal-operation case that CAP ignores.
• Say the classification aloud, for example "DynamoDB is PA/EL," to sound precise.
• Tie the Else branch to a concrete number: quorum writes add a network round trip, often single-digit milliseconds within a region and far more across regions.
• Note that the same database can be tuned per request, so the label describes defaults, not a hard limit.