GUIDES
Decision Engine Examples
Sample aether decide --explain style traces for common workload shapes.
ML inference (GPU, low latency)¶
Spec hints: requirements.gpu: "1", intent.goal: low-latency
| Runtime | Typical outcome |
|---|---|
| KubeVirt | + GPU passthrough, + VM isolation |
| Kubernetes | − GPU scheduling complexity unless device plugin configured |
| Podman | − No production GPU path for large models |
Recommended: KubeVirt.
Cost-optimized API¶
Spec hints: small CPU/memory, intent.goal: cost-optimized, intent.budget.maxMonthlyUsd: 200
| Runtime | Typical outcome |
|---|---|
| Podman | + Zero cloud spend for dev/staging |
| Kubernetes | + Efficient bin-packing at scale |
| KubeVirt | − VM overhead vs containers |
Recommended: Podman (local) or Kubernetes (shared cluster).
HA web service¶
Spec hints: scaling.minReplicas: 3, health probes, intent.resilience: High
| Runtime | Typical outcome |
|---|---|
| Kubernetes | + Native HA, + HPA, + Intent HA bonus |
| KubeVirt | + Strong isolation, − Higher overhead |
| Podman | − Single-node unless orchestrated externally |
Recommended: Kubernetes.
Stateful database (Postgres)¶
Spec hints: large storage, persistence block
| Runtime | Typical outcome |
|---|---|
| Kubernetes | + PVC, snapshots via backup module |
| KubeVirt | + VM + disk workflows |
| Podman | + Simple local dev |
Note: Scoring picks a runtime; data migration is operator responsibility. See Stateful Portability.
Try it¶
aether decide --spec examples/workload-full-featured.yaml --explain
aether intent --spec examples/workload-full-featured.yaml
aether compare --spec examples/workload-full-featured.yaml