The PACELC Extension to CAP
You'll learn how PACELC completes the picture CAP started by addressing trade-offs that exist even when your network is healthy.
Beyond CAP: The Missing Half
CAP tells you what to choose during a partition: consistency or availability. But what about the 99.9% of the time when there's no partition? Your system still faces critical trade-offs.
PACELC extends CAP with this logic:
If there's a Partition (P), choose between Availability (A) and Consistency (C)
Else (E), choose between Latency (L) and Consistency (C)
Why Latency vs Consistency Matters
Even without network failures, distributed systems must decide: should we wait for all replicas to acknowledge a write (strong consistency) or respond faster using cached/local data (better latency)?
The Trade-off in Normal Operation
- Prioritize Consistency: Wait for multiple nodes to sync before confirming operations → slower response times
- Prioritize Latency: Serve reads from nearest replica or accept writes immediately → risk of stale data
Analogy: Imagine a library with multiple branches. Do you wait until all branches update their catalogs before telling a customer a book is available (consistent, but slow)? Or do you check just the nearest branch (fast, but might be outdated)?
PACELC in Practice
- PA/EL systems (like Cassandra, DynamoDB): Choose availability during partitions, prioritize low latency normally
- PC/EC systems (like HBase, traditional RDBMSs): Choose consistency during partitions, maintain it even when it costs latency
Key Takeaway: PACELC reminds us that consistency-latency trade-offs exist continuously, not just during failures. Understanding where your system falls on both the PA/PC and EL/EC spectrums is essential for distributed system design.