21.10 Workload & Performance Architecture
Enterprise performance requires workload-aware architecture.
Workload classes
| Workload | Typical concern |
|---|---|
| Ingestion | throughput and scheduling |
| Transformation | execution time and concurrency |
| Interactive analytics | latency |
| Large scans | data volume and pruning |
| Federation | source latency and pushdown |
| Shared consumption | concurrency and predictable response |
Architecture levers
- reduce unnecessary data movement
- push computation toward the most appropriate execution engine
- persist selectively
- reduce unnecessary columns and rows
- avoid repeated expensive calculations
- separate conflicting workloads when justified
- monitor concurrency and capacity
Capacity model
Required capacity = baseline + growth + peak + safety margin
Performance investigation
Start with evidence:
User symptom -> workload -> execution -> data volume -> bottleneck -> architectural change
Do not optimize from intuition alone.
Architect decision
For each high-volume workload document expected volume, concurrency, freshness, latency and growth. Then choose the physical strategy.