Skip to content
ROOT
STUB

CAP and PACELC

The pattern: a cocktail-party theorem (CAP — under partition, choose Availability or Consistency) and its more useful cousin (PACELC — also during normal operation, choose Latency or Consistency). Together they’re the frame for reasoning about distributed-storage trade-offs.

The trade-off the theorems express: on Partition, A or C + Else, Latency or Consistency. PACELC is more honest because partitions are rare; latency vs. consistency is the everyday trade-off. “We’re partition-tolerant” is meaningless without specifying behavior; “PA/EL” (DynamoDB shape) and “PC/EC” (Spanner shape) are real positions.

[Deepen Year 2 Phase 8 by mapping 5 systems you know to PACELC quadrants.]

  • Replication — every database picks its CAP/PACELC posture inside its replication design.
  • Eventual consistency — the “A” and “EL” half of the trade-off, made operational.
  • Consensus — the “C” and “EC” half, with quorum cost attached.
  • Distributed time — TrueTime is what lets Spanner sit at PC/EC without paying every-write coordination.
  • OLTP vs. OLAP — workload class often dictates which PACELC quadrant is acceptable.

First touched in Year 2 Phase 8; a recurring lens through Year 3 data tier choices.