Cassandra excels at write-heavy workloads:
Architecture:
- Masterless (any node accepts writes)
- Consistent hashing for distribution
- LSM trees for fast writes
Data modeling rules:
- Design tables around queries (one table per query)
- Partition key determines data distribution
- Clustering columns determine sort order within partition
- Denormalize aggressively
Good for:
- Time-series at massive scale
- Write-heavy applications
- Multi-datacenter deployments
Avoid for:
- Ad-hoc queries
- Strong consistency requirements
- Small datasets (overhead not worth it)