Skip to main content

PCNE Certification Preparation Guide: Section 4 — Configuring and implementing hybrid and multicloud network interconnectivity (~16% of the exam)

Honest framing up front: the RAD foundation modules implement none of this section. There is no Interconnect, no VPN, no BGP session, and no NCC hub anywhere in Services_GCP, App_CloudRun, App_GKE, or App_Common. What the platform does give you is a realistic cloud-side anchor — a custom VPC (vpc-network-{prefix}), a Cloud Router (ASN 64514), and a PSA peering with custom-route export — against which every hybrid pattern in this section can be practiced in a scratch project. Deploy the VPC Foundation profile so those anchors exist, then treat this guide as a structured study program. Expect ~16% of the exam from material you will not see running in RAD.

4.1 Configuring Cloud Interconnect​

⏱ ~60 min study · 💰 none (Interconnect cannot be meaningfully lab'd without a circuit) · ⚙️ Requires: VPC Foundation profile for the target VPC only

Why the exam cares — Dedicated vs Partner Interconnect selection (capacity, colocation presence, L2 vs L3 Partner models), VLAN attachment configuration, the 99.9% vs 99.99% SLA topologies, Cross-Cloud Interconnect to other clouds, and encrypting Interconnect with HA VPN over Interconnect or MACsec.

How RAD implements it — Not implemented by the foundation modules.

Try it

  1. You can rehearse the cloud-side objects without a physical circuit (attachments in a scratch project remain unprovisioned but show the workflow):

    gcloud compute interconnects locations list --format="table(name,city,availabilityZone)"
    gcloud compute interconnects attachments partner create my-attachment \
      --region=us-central1 \
      --router=vpc-network-<prefix>-nat-gw-us-central1 \
      --edge-availability-domain=availability-domain-1
    gcloud compute interconnects attachments describe my-attachment \
      --region=us-central1 --format="yaml(pairingKey,state)"

  2. Note the pairingKey — the token you hand to a Partner Interconnect provider — and the PENDING_PARTNER state.

  3. You know you understand it when you can say why the attachment references the Cloud Router (BGP termination) and what changes for a Dedicated attachment (you specify the --interconnect instead of getting a pairing key). Delete the attachment afterward to avoid charges.

Check yourself

Q1: 5 Gbps needed, no presence in a colocation facility, and traffic must be encrypted. Design?

A: Partner Interconnect (no colo presence rules out Dedicated, which starts at 10 Gbps physical circuits) with HA VPN over Interconnect for encryption — Interconnect itself is unencrypted, and MACsec availability depends on the connection type/location. For an L3 partner, the partner's router peers with Cloud Router on your behalf; for L2, you run BGP to Cloud Router yourself.

Q2: What exactly earns the 99.99% Interconnect SLA?

A: Four VLAN attachments on at least two Dedicated/Partner connections in two metros, attachments spread across both edge availability domains in each metro, Cloud Routers in at least two regions, and global dynamic routing mode — plus on-prem redundancy. Two attachments in one metro across both availability domains gets only 99.9%.

Beyond the modules — Study "Cloud Interconnect overview", "Partner Interconnect provisioning", "Cross-Cloud Interconnect" (Google-managed dedicated links to AWS/Azure, same VLAN-attachment + Cloud Router model), "HA VPN over Cloud Interconnect" (VPN gateways on the attachments, doubles as the encryption answer), and MACsec for Cloud Interconnect. Memorize: Dedicated = 10/100 Gbps physical, your colo; Partner = 50 Mbps–50 Gbps via provider; attachments are regional and bind to a Cloud Router.

⚠️ Exam trap — Dataplane: an Interconnect connection is physical and metro-scoped; the VLAN attachment is the regional, routed object. A connection in Chicago can serve attachments to routers in any region (egress costs differ), but SLA math counts metros and availability domains, not regions alone.

4.2 Configuring a site-to-site IPSec VPN​

⏱ ~60 min hands-on possible in a scratch project · 💰 ~$0.05/h per tunnel + egress · ⚙️ Requires: VPC Foundation profile as one side

Why the exam cares — HA VPN (two interfaces, 99.99% with correct tunnel topology, BGP-only) vs Classic VPN (single interface, 99.9%, supports static policy/route-based tunnels), VPN between two VPCs, and interaction with dynamic routing mode.

How RAD implements it — Not implemented by the foundation modules. The deployed Cloud Router ({net}-nat-gw-{region}, ASN 64514) is technically capable of hosting VPN BGP sessions, and the VPC's subnets plus the PSA range (custom-route export already enabled on the peering) are exactly what you would advertise to a remote site.

Try it

  1. This one you can fully lab: create a second VPC in a scratch project and build HA VPN between it and the RAD VPC:

    # One HA VPN gateway per side (note: two interfaces each, automatically)
    gcloud compute vpn-gateways create rad-side-gw --network=vpc-network-<prefix> --region=us-central1
    gcloud compute vpn-gateways create remote-side-gw --network=scratch-vpc --region=us-central1

    gcloud compute routers create remote-router --network=scratch-vpc --region=us-central1 --asn=65010

    # Tunnels (repeat with interface 1 / peer counterpart for full HA)
    gcloud compute vpn-tunnels create rad-to-remote-0 \
      --region=us-central1 --vpn-gateway=rad-side-gw --interface=0 \
      --peer-gcp-gateway=remote-side-gw --shared-secret=SECRET \
      --router=vpc-network-<prefix>-nat-gw-us-central1 --ike-version=2

    gcloud compute routers add-interface vpc-network-<prefix>-nat-gw-us-central1 \
      --interface-name=if-tun0 --vpn-tunnel=rad-to-remote-0 \
      --ip-address=169.254.0.1 --mask-length=30 --region=us-central1
    gcloud compute routers add-bgp-peer vpc-network-<prefix>-nat-gw-us-central1 \
      --peer-name=remote-peer-0 --interface=if-tun0 \
      --peer-ip-address=169.254.0.2 --peer-asn=65010 --region=us-central1

  2. Verify: gcloud compute vpn-tunnels describe rad-to-remote-0 --region=us-central1 --format="value(status)" → ESTABLISHED, then check learned routes with gcloud compute routers get-status.

  3. You know it worked when a VM (or the NFS server VM, tag nfsserver) in the RAD VPC can ping a scratch-VPC VM through the tunnel — remember the intra-VPC firewall rules only allow internal CIDRs, so add an ingress allow for the remote CIDR first.

Check yourself

Q1: An on-prem device supports only policy-based VPN with static routing. HA VPN or Classic?

A: Classic VPN — HA VPN requires BGP. Policy-based/route-based static tunnels exist only on Classic VPN (99.9% SLA, deprecated for new dynamic deployments). The better exam answer when the device can do BGP is always HA VPN with two tunnels for 99.99%.

Q2: HA VPN is up but on-prem can't reach Cloud SQL's private IP, while VMs can. Why?

A: The Cloud SQL instance lives in the producer VPC behind PSA peering. Peering routes aren't advertised over BGP unless the consumer exports them: enable custom-route export on the PSA peering (RAD already does) and add the PSA /16 to Cloud Router's custom advertised routes — the PSA range is not a subnet of the consumer VPC, so default advertisement misses it.

Beyond the modules — Study "HA VPN topologies" (GCP↔GCP, GCP↔on-prem with 2 or 4 tunnels, active/active vs active/passive and the bandwidth-halving caveat), IKE ciphers, and tunnel troubleshooting (Section 5.2). Know link-local BGP addressing (169.254.x.x/30 per tunnel interface, as in the commands above).

⚠️ Exam trap — Creating two tunnels from one HA VPN gateway interface to the peer doesn't earn 99.99% — the SLA requires tunnels from both interfaces of the HA VPN gateway, matched to redundant peer endpoints.

4.3 Configuring Cloud Router​

⏱ ~30 min · 💰 none · ⚙️ Requires: VPC Foundation profile

Why the exam cares — Cloud Router is the BGP speaker behind every dynamic hybrid topology: ASN choice, MED (advertised route priority), custom advertised routes, learned-route priority (--advertised-route-priority, base-priority adjustment), BFD for fast failover, MD5 auth, and best-path selection mode.

How RAD implements it — Partially. The platform creates a real Cloud Router named {net}-nat-gw-{region} with ASN 64514 and no advertised groups, whose only consumer is the Cloud NAT gateway (NAT applied to all subnetworks and all IP ranges). The inline path creates a plain Cloud Router with no BGP configuration at all. No BGP peers, no custom advertisements, no BFD exist anywhere.

Try it

  1. Inspect the live router and add a custom advertisement (harmless with no peers — it changes what would be advertised):

    gcloud compute routers describe vpc-network-<prefix>-nat-gw-us-central1 \
      --region=us-central1 --format="yaml(bgp,nats[].name)"
    gcloud compute routers update vpc-network-<prefix>-nat-gw-us-central1 \
      --region=us-central1 \
      --advertisement-mode=CUSTOM \
      --set-advertisement-groups=ALL_SUBNETS \
      --set-advertisement-ranges=<psa-range-cidr>=PSA-range
    gcloud compute routers get-status vpc-network-<prefix>-nat-gw-us-central1 \
      --region=us-central1

  2. You know it worked when describe shows advertiseMode: CUSTOM with your range. Revert to --advertisement-mode=DEFAULT afterward to avoid Terraform drift on the next platform apply.

Check yourself

Q1: Two Interconnect attachments; you want one preferred for traffic to on-prem and on-prem to prefer one path back. Which knobs?

A: Inbound to Google: on-prem influences Google's choice via MED it sends; Google-side preference for learned identical prefixes follows the route's priority (derived from MED + inter-regional cost). Outbound from Google: set --advertised-route-priority (MED Google sends) per BGP peer — lower MED = more preferred by on-prem. Asymmetry questions almost always resolve to "MED in each direction".

Q2: Failover between two VPN tunnels takes ~60s. How do you make it sub-second?

A: Enable BFD on both BGP peers (gcloud compute routers update-bgp-peer --bfd-session-initialization-mode=ACTIVE --bfd-min-transmit-interval=...). BGP hold timers alone are tens of seconds; BFD detects dataplane failure in hundreds of milliseconds and tears the route down immediately.

Beyond the modules — Study "Cloud Router overview": ASN rules (private 64512–65534/4200000000+ ranges; Google's side of PSA-style peering vs your --asn), regional vs global dynamic routing and how it changes which subnets the router advertises, MD5 authentication on BGP sessions, and legacy vs standard best-path selection modes. Also note each NAT-only router (RAD's case) still counts against router quotas.

⚠️ Exam trap — Cloud Router advertises subnet routes (per routing mode) by default; custom routes, peering ranges (like RAD's PSA /16), and secondary ranges outside the mode's scope require CUSTOM advertisement mode. "It's in the VPC so it's advertised" fails for PSA ranges.

4.4 Configuring Network Connectivity Center​

⏱ ~30 min study · 💰 none · ⚙️ Requires: nothing — concept-only

Why the exam cares — Section 4's NCC angle is the hybrid one (vs Section 2.3's VPC-spoke angle): VPN/Interconnect attachments as hybrid spokes, site-to-site data transfer through Google's backbone, router appliances (SD-WAN integration) as spokes peering BGP with Cloud Router, and the transitivity rules.

How RAD implements it — Not implemented by the foundation modules.

Try it

  1. If you built the 4.2 HA VPN lab, promote it into an NCC topology in the scratch project:

    gcloud network-connectivity hubs create hybrid-hub
    gcloud network-connectivity spokes linked-vpn-tunnels create vpn-spoke \
      --hub=hybrid-hub --region=us-central1 \
      --vpn-tunnels=rad-to-remote-0,rad-to-remote-1 \
      --site-to-site-data-transfer
    gcloud network-connectivity spokes list --hub=hybrid-hub

  2. You know it worked when the spoke shows ACTIVE and the hub's route table includes prefixes learned from the tunnels.

Check yourself

Q1: Two branch offices, each VPN'd to GCP, need branch-to-branch traffic without new circuits. Solution?

A: NCC hub with both VPN tunnel sets as hybrid spokes and site-to-site data transfer enabled — branch traffic transits Google's backbone between the spokes. Without NCC, two VPN tunnels into one VPC do not forward traffic between each other (a VPC is not a transit router for external-to-external flows).

Q2: Where do router appliances fit?

A: A router appliance spoke is a VM (typically a vendor SD-WAN appliance) in the VPC that BGP-peers with Cloud Router; NCC then treats its learned prefixes like any hybrid spoke. It's the answer pattern for "integrate our SD-WAN fabric with Google Cloud".

Beyond the modules — Study "NCC site-to-site data transfer" (supported regions, billing), router-appliance BGP setup, mixing VPC spokes with hybrid spokes (hub provides the transitivity that peering lacks — but VPC spokes and hybrid spokes interoperate per documented rules, not unconditionally), and Private NAT at the hub for overlapping site ranges.

⚠️ Exam trap — "VPC peering + VPN = on-prem reaches the peered VPC" is false (non-transitive). The two supported fixes are: export/import custom routes across the peering with Cloud Router advertising them, or restructure with NCC spokes. Recognize which the scenario allows.