Substrate Architecture #

Within a site, the substrate provides the shared infrastructure foundation — networking, compute, and services, all defined as code. The substrate is stateless — all configuration lives in source control and is applied through automation, so it can be reprovisioned from scratch at any time.

For site definitions and the independence model, see Architecture.

Infrastructure Layers #

block-beta
    columns 2
    hdr["Substrate Infrastructure"]:2
    net["Network"]:1 netd["Routing · Firewall · DNS · DHCP · NAT · Switching · Wireless"]:1
    cmp["Compute · Storage¹"]:1 cmpd["Hypervisors · Persistent storage"]:1
    mgp["Management Plane"]:1 mgpd["Network management · Substrate observability"]:1
    ctp["Control Plane"]:1 ctpd["Provisioning API · Tenant DNS · Secrets · Broker"]:1

¹ Storage is each hypervisor’s local disks. Shared storage is not planned.

Network #

Connectivity, segmentation, and the foundational network services. Nothing else in the substrate has to be running for these to work:

Routing and gateway: NAT, inter-segment routing, upstream connectivity
Firewall: segment isolation and egress policy
DNS: authoritative resolution for the substrate zone
DHCP: static mappings for known hosts, dynamic pools per segment
Switching and wireless: VLAN trunking and wireless access

See Networking for substrate networking services. See Network Segmentation for the segment model and trust hierarchy.

Compute and Storage #

Compute is the virtualization hosts: a management hypervisor carrying both planes’ VMs, and a tenant hypervisor carrying the fabric and what runs on it. See Compute.

Storage¹ is shared and persistent storage for substrate consumers. See Storage.

Management Plane #

What the substrate runs so it can manage and observe itself — network device management and substrate observability, on the management segment, for operators and substrate hosts. See Management Plane.

Control Plane #

What the substrate runs so it can serve what runs on it — the provisioning API that creates tenants, tenant DNS, the secret store, tenant observability, and the device broker. It sits on the Platform and IoT Backend segments, because tenants and devices must never reach the management segment. See Control Plane.

The two planes are separated by audience, not by technology, and no host belongs to both. That separation is what lets a tenant depend on the substrate without the substrate coming to contain the tenant.

Child Documents #

Networking — Networking services: DNS, DHCP, firewall, VLAN routing, switching
Compute — Virtualization and compute model
Storage — Shared and persistent storage
Management Plane — How the substrate manages and observes itself
Control Plane — How the substrate serves the tenants and devices on it