Operational Scale: Vertical vs Horizontal Growth

Vertical Scaling (Relational)
Relational databases scale up by adding CPU, memory, and faster storage. A single node handles 10K-50K TPS with sub-millisecond latency. Beyond this, scaling out requires sharding (splitting data by key), where the application routes queries to the correct shard.
Horizontal Scaling (NoSQL)
NoSQL scales horizontally from the start. Adding nodes linearly increases capacity. Clusters handle millions of ops/sec. However, hot partitions cap per-key throughput at 1K-5K ops/sec. Write amplification: 1 logical write becomes 5-10 physical writes from indexes and replication.
Scale Comparison
50K
TPS/NODE (SQL)
1M+
TPS CLUSTER (NoSQL)
Cost Trade-offs
Relational delivers higher per-node performance. NoSQL inflates storage 2-5x from replication and denormalization. Choose relational for <50K TPS with complex queries; NoSQL for millions of ops/sec with defined access patterns.

💡 Key Takeaways

✓Relational scales vertically to 10K-50K TPS per node; beyond requires explicit sharding or distributed SQL

✓NoSQL scales horizontally by adding nodes linearly, but hot partitions cap per-key throughput at 1K-5K ops/sec

✓Write amplification in NoSQL: 1 logical write becomes 5-10 physical writes from indexes, denormalization, and replication

✓Storage cost trade-off: NoSQL inflates storage 2-5x due to replication and denormalization

📌 Interview Tips

1Quantify scale thresholds: single relational node handles 50K TPS; beyond that, consider sharding or distributed SQL

2Explain hot partitions: even with millions of cluster TPS, a single key is limited to a few thousand ops/sec on its partition

3For cost discussions, mention that NoSQL requires more storage (2-5x) but provides linear horizontal scalability

← Back to Relational vs NoSQL Overview