Tenant Compute

Tenant Compute #

The tenant compute layer provides resources for application workloads, experiments, and user-facing services. Unlike the management plane, tenant compute is designed for higher churn and experimentation.


Tenant Compute Components #

ComponentPurpose
Tenant HypervisorsProxmox hosts for VM-based tenant workloads
Raspberry PIsEdge/IoT compute for specialized workloads

Characteristics #

Tenant compute infrastructure is:

  • Tolerant of rebuilds — Expect frequent provisioning and teardown
  • Fast change cadence — Experimentation and iteration encouraged
  • Terraform-managed — Declarative VM lifecycle (future)
  • VLAN-isolated — Tenant traffic separated from management plane

Workload Types #

CategoryExamples
Application developmentIoT backend, services, APIs
ExperimentsTest environments, sandboxes
Ephemeral workloadsShort-lived or rebuildable VMs
Edge/IoTSensor collection, local processing

Separation from Management Plane #

Tenant compute is deliberately separated from management plane infrastructure:

AspectManagement PlaneTenant Compute
Change cadenceSlow, deliberateFast, experimental
Blast radiusMust be minimizedTolerable
Rebuild toleranceLow — avoid rebuildsHigh — expect rebuilds
ProvisioningAnsibleTerraform (future)

This separation ensures that tenant experimentation cannot impact substrate stability. If a tenant VM misbehaves or a tenant hypervisor fails, the management plane remains operational.


Tenant Networking #

Tenant networks are EVPN/VXLAN overlays owned by the tenant hypervisor’s own SDN fabric — one VRF per tenant, an anycast gateway hosted by the fabric, and addressing from fabric IPAM. The core router is the perimeter for the aggregate transit network only.

See ADR-0001 and the tenant fabric implementation.

Page last modified: August 30, 2026