Vivcre Learn learn it · write it · retain it

← All posts

System Design #pacelc#consistency#latency#eventual-consistency#distributed-systems

PACELC: The Tradeoff That's Always On

9 Jul 2026

CAP describes what happens during a partition — a rare event. But there’s a second dial that’s turned even when the network is perfectly healthy, and it governs every read you ever serve.

The always-on dial

Picture the system running normally, cables all intact. A write to replica A still takes some time to reach replica B. During that gap, a read at B is stale. So on every single read there’s a standing question: how long are you willing to wait to be sure the read reflects the latest write? Confirm with A on every read and reads are slow but always fresh; answer immediately from local state and reads are fast but possibly stale.

That’s the tradeoff PACELC names. Read it as:

if Partition, then Availability or Consistency; Else, Latency or Consistency.

CAP is only the “P” half. The “ELC” half is the everyday truth: with a healthy network you are always trading Latency vs Consistency. An analytics dashboard tolerates staleness to answer fast (EL — favor latency); a bank balance pays the latency to be exact (EC — favor consistency).

Eventual consistency

Eventual consistency is the promise an EL/AP system makes: if writes stop, all replicas will converge to the same value — eventually — but at any given instant a read may be stale. The word doing the work is “eventually”: there’s no bound on when, only a guarantee that it will. That’s what lets a cart or a feed answer instantly and reconcile in the background.

It’s a spectrum, not two points

Between “always fresh” (strong) and “eventually, whenever” (eventual) sit useful middle guarantees:

These exist because strong is expensive and eventual is sometimes too weak, so systems expose tunable knobs. Cassandra literally lets you pick consistency per query — ONE for fast/stale, QUORUM for slower/fresher — which is PACELC’s EL-vs-EC dial handed directly to you, the engineer.

The whole distributed-data picture at the tradeoff level: CAP for the partition case, PACELC for the everyday case, and the consistency spectrum in between — all the one instinct carried the whole way: what will you sacrifice, and when?


Practise these questions →

Spaced-repetition MCQs for this post, on practise.vivcre.com.