Beyond CAP: The Need for PACELC
What you'll learn: Why CAP theorem alone is incomplete and how PACELC fills the gap by addressing what happens when your system isn't experiencing a partition.
The Problem with CAP Alone
CAP theorem tells us what to choose during network partitions: consistency or availability. But here's the catch—partitions are rare. Your system spends most of its time operating normally, without partitions. CAP says nothing about this normal state, yet critical design decisions still must be made.
Enter PACELC
PACELC extends CAP to cover both scenarios:
"If there is a Partition (P), choose between Availability and Consistency (AC); Else (E), when the system is running normally, choose between Latency and Consistency (LC)."
The key insight: even without partitions, you're still making tradeoffs. Every replica update creates a choice:
- Wait for all replicas to confirm (strong consistency, higher latency)
- Return immediately (lower latency, eventual consistency)
Why This Matters
Imagine a distributed database serving normal traffic. No network partition exists. Do you:
- Synchronously replicate writes to all nodes? (Consistent but slower)
- Asynchronously replicate in the background? (Faster but temporarily inconsistent)
CAP doesn't help here. PACELC does.
Real example: Cassandra is PA/EL (available during partitions, prioritizes low latency during normal operation). DynamoDB with strong consistency is PC/EC (consistent during partitions, consistent but higher-latency normally).
PACELC acknowledges that latency is always part of the equation, not just an afterthought.
Key Takeaway: CAP only describes partition behavior; PACELC reminds us that normal operation requires choosing between latency and consistency, making it a more complete framework for real-world system design.