Reference data

Internal CIDRs — Reference

Several VCF 9.1 components carve an internal, non-routed address block out of the IP space for their own containers, pods, services or transit links. None of these blocks needs a VLAN, a gateway or a DNS record, so they are easy to forget in the Step 1 plan (01-network-dns-plan.md). But each one is still a route inside the component, so if it overlaps a network you actually route, traffic to that network is silently sent inward and never leaves.

This page lists all of them in one place: the default, the alternatives you can pick, where the value is set, and whether you can change it later.

The one rule: no internal block may overlap anything you route (management, VM, vMotion, vSAN, TEP, edge uplinks, VPC external blocks, user and branch-office networks, VPN ranges). Unless a row below says sharing is allowed, they should not overlap each other either.

Evidence level: TechDocs / KB / API reference sourced, links per row, checked 2026-10-22 against the VCF 9.1 / 9.1.1 pages. The License Hub 2.0 and Cloud Proxy rows are also field-verified (see 15-license-hub.md and 27-vcf-operations-ha-cloud-proxy-vvf.md).


1. Summary table

#ComponentDefaultAlternativesWhere it’s setChange later?
1VCF services runtime (VCF Management Services, VCF and VVF)198.18.0.0/15240.0.0.0/15 or 250.0.0.0/15, fixed list. IPv6: fd00::/111 in the API sample9.1.0: JSON spec / SDDC Manager API only (internalClusterCidrIpv4). 9.1.1+: VCF Installer UI, Internal cluster CIDR IPv4Not documented; treat it as fixed at deploy
2VCF Automation services runtime198.18.0.0/15240.0.0.0/15 or 250.0.0.0/15. May share row 1’s block9.1.0: fleet lifecycle API only. 9.1.1+: VCF Operations UINot documented; treat it as fixed at deploy
3License Hub 2.0 Internal Cluster Network10.10.0.0/16No fixed list from Broadcom, only a rule: unused, non-routable, ≥512 addresses (/23 or larger). Suggested: 10.255.0.0/16, 192.168.254.0/23 (see §2.2)OVA deploy, Customize templateNot documented; FQDNs and IPs are immutable, so assume this is too
4vDefend SSP pod CIDR (SSP Installer flow)Not stated by BroadcomAny unused block, e.g. 10.27.0.0/16 (KB example)secop_capi_pod_cidr in /config/clusterctl/settings.json on the SSP Installer appliance, before deployingNo. Changing it afterwards “will not affect existing cluster”
5Cloud Proxy Docker subnetDocker’s own: 172.17.0.0/16 (bridge), then 172.18.0.0/16, …Set it explicitly, /27 or largerOVA deploy, Docker Subnet fieldNo. Only by deleting and recreating the Docker networks
6Supervisor Service CIDRPre-filled by the wizard; no default documentedAny private block; this repo’s example is 10.96.0.0/23. VDS networking: /16. Classic NSX: /16 minimumSupervisor activation wizard / API (default_service_cidr in the 9.1 API)Not without redeploying the Supervisor
7Supervisor Pod CIDR (classic NSX segment networking only, API-only in 9.1)None/23 minimum; this repo’s example is 10.244.0.0/20Supervisor JSON / APINot without redeploying the Supervisor
8VKS guest cluster Pod / Service CIDRsNone: “required, there is no default”Broadcom’s examples: pods 192.168.0.0/16, services 10.96.0.0/12. Per clusterCluster manifest, spec.clusterNetworkNot on a running cluster
9NSX Tier-0 ↔ Tier-1 transit100.64.0.0/16Any unused blockTier-0 transit_subnetsYes, via the Policy API (see §2.6)
10NSX SR ↔ DR internal transit169.254.0.0/24 (active-active), 169.254.0.0/28 (active-standby)IPv4 onlyTier-0 internal_transit_subnetsRarely needed; link-local
11NSX default T0 ↔ child VRF transit169.254.2.0/23—Tier-0 vrf_transit_subnetsRarely needed; link-local
12NSX T0/VRF ↔ Transit Gateway transit169.254.4.0/22—Tier-0 tgw_transit_subnetsRarely needed; link-local
13NSX DHCP (reserved)100.96.0.0/30Cannot be moved: keep every other range off itCreated automatically when DHCP runs on a Tier-1—

Not in this table:

  • Routed blocks that sit next to these in the wizards: the services-runtime node IP pool and the VCF Automation nodes CIDR (both on VM Management, see 01-network-dns-plan.md), the VPC External IP Blocks, and Supervisor ingress/egress (10-supervisor-enablement.md).
  • VPC Private CIDRs and the Private Transit Gateway IP Block. They don’t leave NSX, but you size and plan them like any other address block: see 10-supervisor-enablement.md §3.3. A /26 of the Private VPC CIDRs is reserved for the Avi Service Engines.

2. Per component

2.1 VCF services runtime and VCF Automation (rows 1–2)

One fixed pick-list for both, verbatim from TechDocs (Deploy VCF Management Services … as Part of VCF Upgrade to 9.1):

“VCF services runtime uses an internal CIDR block of 198.18.0.0/15. If this subnet is in use on your network, you must change the internal CIDR block for VCF management services to 240.0.0.0/15 or 250.0.0.0/15.”

Sharing is allowed. The same page: “Different VCF services runtime instances on the same network can work with the same internal network. For example, you can have the two instances of VCF services runtime for your VCF management services and VCF Automation using the same internal network.” So leaving both at 198.18.0.0/15 is fine. Move both only if that block clashes with something routed.

UI vs API depends on the version. “With the VCF 9.1.0.x UI, you cannot change the default internal network. With VCF 9.1.1 and later, you can change the default internal network.” On 9.1.0 you set it in the JSON spec (VCF Installer) or the SDDC Manager API for VCF Management Services, and in the fleet lifecycle API for VCF Automation. “Starting with VCF 9.1.1 you can specify the internal networks during the deployment procedures by using the VCF Installer UI or VCF Operations UI” (First VCF Instance FQDNs and IP addresses). The 9.1.1 field help: “Select a CIDR block that does not overlap with any of your infrastructure networks, including management, VM, and IP pool networks. An overlap causes routing conflicts and breaks connectivity to or from the VCF services runtime cluster.”

The API sample uses "internalClusterCidrIpv4": "198.18.0.0/15" and, commented out, "internalClusterCidrIpv6": "fd00::/111".

Log Management and Real-time Metrics run on the services runtime, so they have no internal block of their own. The Internal cluster CIDR in 31-log-management.md is row 1, set when VCF Management Services is added to a VVF deployment.

Picking an alternative: 198.18.0.0/15 is the IANA benchmarking range, and 240.0.0.0/15 / 250.0.0.0/15 sit in the reserved class E space, so none of them should normally be routed. Check for 198.18.0.0/15 in lab and test-equipment networks: that’s where it is most often used.

2.2 License Hub 2.0 (row 3)

The OVA field defaults to 10.10.0.0/16, and its tooltip (field-verified 2026-08-24) reads: “The CIDR must support at least 512 IP addresses (prefix /23 or smaller). Use non-routable IPs as this is for internal use only, and ensure the range is not used elsewhere in your datacenter. Most deployments can leave this at the default value.” (License Hub Appliance System Requirements).

10.10.0.0/16 sits in RFC 1918 space, so it collides more often than the services-runtime defaults. Check it against your IP plan.

Alternatives. Unlike the services runtime, License Hub has no fixed pick-list: the tooltip above is the only guidance. Neither the system requirements page nor the API reference lists allowed values. The options below are this repo’s suggestions, not a Broadcom list. Try them in this order:

OptionBlockNotes
1. Keep the default10.10.0.0/16Broadcom’s own advice, if nothing in your estate uses it
2. Another unused RFC 1918 blocke.g. 10.255.0.0/16, or a /23 such as 192.168.254.0/23The top of 10.0.0.0/8 is rarely allocated. A /23 is the minimum and is enough: you only need room for the appliance’s own pods and services
3. A smaller unused slice you reserve in IPAMany free /23–/16Register it in IPAM as “License Hub internal, do not route” so nobody allocates it later

Avoid these:

  • 198.18.0.0/15, 240.0.0.0/15, 250.0.0.0/15 (the services-runtime list). Broadcom only allows sharing between services-runtime instances, not with License Hub, and it hasn’t been verified whether the License Hub OVA accepts the class E blocks (240/250.x).
  • 172.17.0.0/16–172.31.0.0/16. These are Docker’s default pools (row 5). Keep them free so container-based appliances don’t collide with them.

Broadcom doesn’t document changing it after deployment, so decide it up front and record it in the Step 1 plan. See 15-license-hub.md.

2.3 vDefend SSP (row 4)

KB 388890 (vDefend 5.0), verbatim: “In some customer environment, the default pod CIDR might cause conflict. An option is given to override the pod CIDR range.” Before deploying SSP, SSH to the SSP Installer appliance as root and add "secop_capi_pod_cidr": "<desired CIDR>" (example 10.27.0.0/16) to /config/clusterctl/settings.json. “Changing the CIDR range after deployment will not affect existing cluster.”

The KB doesn’t name the default pod CIDR. It doesn’t mention a service CIDR either. It is written for vDefend 5.0. License Hub 5.1.2 deploys through the same SSP Installer, so the override probably applies there too. That is not verified. See 18-vdefend-ssp.md.

2.4 Cloud Proxy Docker subnet (row 5)

KB 392302: “The 172.17.0.0/16 subnet is used for the default bridge network”, and for user-defined bridge networks “Docker assigns a different subnet by incrementing the second byte (e.g., 172.18.0.0/16)”. “If a Docker network overlaps with the external environment network, connectivity problems may occur, as network packets won’t be routed outside the Cloud Proxy but will instead be routed internally.” The only fix afterwards is to delete and recreate the Docker networks. So set the Docker Subnet field (/27 or larger) at deploy time, in a block you know is free. See 27-vcf-operations-ha-cloud-proxy-vvf.md.

2.5 Supervisor and VKS (rows 6–8)

  • Supervisor Service CIDR: unique per Supervisor, and it must not overlap the VPC blocks, other Supervisors or any fleet network (10-supervisor-enablement.md §2). The 9.1 API carries it as default_service_cidr. Broadcom doesn’t publish a default value, so take the one the wizard offers only after checking it against your plan.
  • VKS guest clusters: the v1beta1 / v1beta2 Cluster has no default. The VKS 3.5.0 ClusterClass example says “specify the cluster network; required, there is no default” and marks both ranges “CAUTION: must not overlap with Supervisor”.
  • Watch the example values. Broadcom’s guest-cluster service example, 10.96.0.0/12, covers 10.96.0.0–10.111.255.255. That includes a Supervisor Service CIDR of 10.96.0.0/23. If you copy both examples, the ranges overlap. This repo’s examples avoid that (10.97.0.0/16 for guest services in 20-supervisor-image-registry.md), but 10.97.0.0/16 also sits inside 10.96.0.0/12. Pick one scheme and check each guest cluster’s ranges against the Supervisor’s.
  • Proxy no-proxy list: every one of these ranges goes into the Supervisor and VKS no-proxy list (20-supervisor-image-registry.md).

2.6 NSX transit ranges (rows 9–13)

From the NSX 9.1.1 Policy API, Tier-0 schema (Tier0):

  • transit_subnets: “When not specified, subnet 100.64.0.0/16 is configured by default.” Used for the Tier-0 ↔ Tier-1 links.
  • internal_transit_subnets: “Only IPv4 addresses are supported. When not specified, subnet 169.254.0.0/24 is assigned by default in ACTIVE_ACTIVE HA mode or 169.254.0.0/28 in ACTIVE_STANDBY mode.”
  • vrf_transit_subnets: default 169.254.2.0/23 (default T0 ↔ child VRFs).
  • tgw_transit_subnets: default 169.254.4.0/22 (default T0/VRF ↔ Transit Gateway).

The 169.254.x ranges are link-local and never conflict with routed networks. 100.64.0.0/16 is the one that bites. It sits inside the CGNAT range 100.64.0.0/10, which some carriers, SD-WAN and VPN products route. If you route anything in 100.64.0.0/16, set a different transit_subnets on the Tier-0. KB 428847 shows the Policy API GET/PATCH to do it.

Don’t move it onto 100.96.0.0/30. That KB: “The IP range 100.96.0.0/30 is a reserved internal range used by NSX for DHCP services on segments.” A Tier-0 transit_subnets that overlaps it fails with error 500105.


3. Planning checklist

  • Every row in §1 that applies to your scope has a block, written into the Step 1 plan next to the routed subnets.
  • Each block has been checked against the whole routed estate, not only the VCF subnets: user, branch, VPN, partner and cloud ranges.
  • Services runtime + VCF Automation: one shared block from the fixed list (198.18.0.0/15 unless it clashes).
  • License Hub 2.0: 10.10.0.0/16 checked against your RFC 1918 plan.
  • Cloud Proxy: an explicit Docker subnet, for every Cloud Proxy.
  • Supervisor Service CIDR unique per Supervisor. Guest-cluster ranges don’t overlap it.
  • NSX: nothing routed in 100.64.0.0/16, or a different Tier-0 transit_subnets is planned.
  • Everything that can’t be changed later (rows 1–8) is decided before deployment.

ITQ – What's Next. VCF 9.1 deployment-planning and operations helper material. © 2026 Paul van Dieen. hollebollevsan.nl