Comparing Serverless vs Pod-Based
1. Understanding Serverless Pricing
| Component | Billing |
|---|
| Storage | Per GB-month |
| Read units | Per query (RUs) |
| Write units | Per upsert (WUs) |
| Min cost | $0 when idle |
2. Understanding Pod Pricing
| Component | Billing |
|---|
| Pods | Per hour × pod count × size |
| Replicas | Linear multiplier |
| Storage | Included in pod cost |
| Always-on | Pay whether queried or not |
Serverless
- Cold start: ~1–5s after idle
- Warm latency: comparable to pods
- Scales auto with load
- Variable cost
Pod-Based
- Always warm; no cold start
- Predictable p99 latency
- Manual scaling
- Fixed cost
4. Comparing Scaling Behavior
| Aspect | Serverless | Pod |
|---|
| Read scaling | Auto | Add replicas |
| Storage scaling | Auto (∞) | Add shards / pod size |
| Write throughput | Auto-batched | Per pod limit |
5. Choosing for Variable Workloads
Note: Pick serverless for spiky, low-utilization, or unpredictable traffic. Cold starts only hurt the very first query.
| Workload | Choice |
|---|
| RAG dev/test | Serverless |
| Internal tools | Serverless |
| Long-tail SaaS tenants | Serverless |
6. Choosing for Predictable Workloads
| Workload | Choice |
|---|
| Sustained high QPS | Pod (p2) |
| Strict p99 SLA | Pod |
| Large fixed corpus | Pod (s1) or serverless |
7. Understanding Feature Differences
| Feature | Serverless | Pod |
|---|
| Collections | No (uses backups) | Yes |
| Selective metadata indexing | No | Yes |
| Auto-scaling | Yes | Manual |
| Imports | Yes (bulk import) | No |
| Namespaces | 10K+ | ~100K |
8. Comparing Regional Availability
| Cloud | Serverless | Pod |
|---|
| AWS | Multiple regions | Multiple regions |
| GCP | Available | Available |
| Azure | Limited | Limited |
9. Migrating Between Types
Pod → Serverless Migration
- Create serverless index with same dim/metric.
- Export from pod (collection or paginated query+fetch).
- Re-upsert in batches into serverless.
- Dual-write briefly; validate parity.
- Cut over reads; delete pod index.
10. Implementing Hybrid Deployment Strategies
| Pattern | Use Case |
|---|
| Hot pod + cold serverless | Active tenants on pod, archive on serverless |
| Region split | Pod in primary, serverless in secondary |
| Dev vs prod | Serverless dev, pod prod (or vice versa) |