Skip to content

Professional Cloud Network Engineer · Domain 1: Designing and planning a Google Cloud VPC network · 21% of the exam

Task 1.3: Designing a resilient and performant hybrid and multi-cloud network.

Cloud Interconnect, Cloud VPN and SD-WAN for hybrid links, Cross-Cloud Interconnect, Direct Peering and Verified Peering Provider, regional or global dynamic routing for multi-region resilience, reaching several VPC networks and Google APIs from on-premises, address planning, hybrid DNS, MTU and encryption with MACsec or HA VPN over Cloud Interconnect.

Study it

  • Connectivity to managed services: private services access, Private Service Connect, Serverless VPC Access and Direct VPC egress

    Lesson coming

  • IP address planning: subnets, IPv6, BYOIP, PUPI, non-RFC 1918 ranges, internal ranges, Private NAT and MTU

    Lesson coming

  • Hybrid and multicloud design: Cloud Interconnect, Cloud VPN, peering options, dynamic routing mode and hybrid DNS

    Partly covered by: Hybrid Networking & DNS

Sample questions

Try each one before opening the answer. Every option is explained, with the Google Cloud documentation page that proves it.

Question 1 · choose 1

The VPC network prod-vpc already has a Cloud Router in us-east4 whose HA VPN tunnels to the data center use the BGP peering blocks 169.254.21.0/30 and 169.254.21.4/30. The company now adds two HA VPN tunnels from a new HA VPN gateway and Cloud Router in us-central1 of prod-vpc to an active-active VPN gateway in Microsoft Azure, with BGP on both tunnels. Which BGP peering address blocks can the two new tunnels use?

  1. A169.254.1.0/30 and 169.254.1.4/30
  2. B169.254.21.0/30 and 169.254.21.4/30, reused because they belong to a Cloud Router in another region
  3. C169.254.22.0/30 and 169.254.22.4/30
  4. D169.254.22.0/29 and 169.254.22.8/29
Show the answer and why
  • A169.254.1.0/30 and 169.254.1.4/30

    Incorrect

    Azure accepts APIPA BGP peering addresses only from 169.254.21.* and 169.254.22.*.

  • B169.254.21.0/30 and 169.254.21.4/30, reused because they belong to a Cloud Router in another region

    Incorrect

    The BGP peering addresses must be unique among all Cloud Routers in all regions of the VPC network, so the us-east4 blocks can't be reused.

  • C169.254.22.0/30 and 169.254.22.4/30

    Correct

    Both blocks are in an Azure APIPA range, one /30 per tunnel, and no other Cloud Router in prod-vpc uses them.

  • D169.254.22.0/29 and 169.254.22.8/29

    Incorrect

    Each tunnel takes a pair of peering addresses in a /30 block, not a /29.

For HA VPN to an Azure active-active VPN gateway, each tunnel needs a pair of link-local BGP peering addresses in a /30 block taken from 169.254.21.* or 169.254.22.*, the valid ranges for Azure APIPA BGP addresses. On the Google Cloud side, the addresses must also be unique among all Cloud Routers in all regions of the VPC network. With the 169.254.21.0/30 and 169.254.21.4/30 blocks taken in us-east4, unused /30 blocks such as those in 169.254.22.* are left.

Question 2 · choose 1

A VPC network uses global dynamic routing. Its only Cloud Routers are in us-east4 and europe-west1, each with BGP sessions over VLAN attachments in its own region, and both learn the data center prefix 10.50.0.0/16. Application VMs run in us-east4, europe-west1 and us-central1. To keep European traffic on the European attachments, a colleague proposes switching the network to regional dynamic routing. What happens to the VMs in us-central1 after the switch?

  1. AThey have no dynamic route to 10.50.0.0/16 anymore, because no Cloud Router in their own region learns the prefix
  2. BThey keep reaching 10.50.0.0/16 through the us-east4 attachments, the region with the lower inter-regional cost
  3. CThey spread traffic to 10.50.0.0/16 across the attachments in both regions, because both routes have equal MED values
  4. DThey keep their routes, because the dynamic routing mode only changes which subnet ranges the Cloud Routers advertise
Show the answer and why
  • AThey have no dynamic route to 10.50.0.0/16 anymore, because no Cloud Router in their own region learns the prefix

    Correct

    In regional mode each region's dynamic routes have next hops only in that region and come only from that region's Cloud Routers; us-central1 has none.

  • BThey keep reaching 10.50.0.0/16 through the us-east4 attachments, the region with the lower inter-regional cost

    Incorrect

    Adding inter-regional cost to routes from other regions is global mode behavior; regional mode does not share routes between regions at all.

  • CThey spread traffic to 10.50.0.0/16 across the attachments in both regions, because both routes have equal MED values

    Incorrect

    Equal MED values matter only among routes a region actually has; after the switch us-central1 receives none of these routes.

  • DThey keep their routes, because the dynamic routing mode only changes which subnet ranges the Cloud Routers advertise

    Incorrect

    The mode controls both directions: which subnets are advertised and in which regions learned routes are applied.

In regional dynamic routing mode, each region's dynamic route control plane processes only the routes learned by Cloud Routers in its own region, and the resulting routes have next hops only in that region. Regions without a Cloud Router, such as us-central1 here, get no dynamic routes to the data center. Global mode shares each region's best path with every other region.

Question 3 · choose 1

A data center reaches the VPC network consumer-vpc only through VLAN attachments in us-west1; the network uses global dynamic routing, and its Cloud Router advertises all subnets. In us-east1, a Private Service Connect endpoint targets a partner's service, published last year behind a regional internal Application Load Balancer whose global access was off when the service attachment was created. VMs in us-east1 reach the service, but on-premises clients time out. Which plan gives the on-premises clients access?

  1. Aconsumer-vpc turns on global access on the existing endpoint, with no change on the partner's side
  2. BThe partner enables global access on its load balancer and publishes a new service attachment, used by a new endpoint with global access
  3. CThe partner turns on global access on its load balancer, and consumer-vpc turns on global access on the existing endpoint
  4. Dconsumer-vpc creates a second endpoint for the same service attachment in us-west1, where the VLAN attachments are
Show the answer and why
  • Aconsumer-vpc turns on global access on the existing endpoint, with no change on the partner's side

    Incorrect

    With a regional internal Application Load Balancer as producer, consumer global access works only if the load balancer had global access before the service attachment was created.

  • BThe partner enables global access on its load balancer and publishes a new service attachment, used by a new endpoint with global access

    Correct

    On-premises traffic enters in us-west1, so the endpoint needs global access, which this producer supports only if its load balancer had global access before the service attachment was created.

  • CThe partner turns on global access on its load balancer, and consumer-vpc turns on global access on the existing endpoint

    Incorrect

    Turning on the load balancer's global access now comes after the service attachment was created, which doesn't meet the condition.

  • Dconsumer-vpc creates a second endpoint for the same service attachment in us-west1, where the VLAN attachments are

    Incorrect

    An endpoint must be created in the same region as the published service that it targets.

By default, a Private Service Connect endpoint can be reached only by clients in its own region and network; global access lets clients in any region, including on-premises clients that enter through VLAN attachments elsewhere, use it. Whether consumer global access works depends on the producer: for a regional internal Application Load Balancer, only if global access was enabled on the load balancer before the service attachment was created. An endpoint must also be in the same region as the published service.

Question 4 · choose 1

A company has Direct Peering with Google at a peering facility. An architect proposes using it so that on-premises servers reach Compute Engine VMs on their internal IP addresses. What should the network engineer point out?

  1. ADirect Peering creates no VPC routes; on its own it reaches only Google's public services
  2. BDirect Peering reaches internal addresses only after a Cloud Router is attached to the peering sessions
  3. CDirect Peering reaches internal addresses only in regions near the peering facility
  4. DDirect Peering reaches internal addresses, but only with Google's 99.99% SLA purchased separately
Show the answer and why
  • ADirect Peering creates no VPC routes; on its own it reaches only Google's public services

    Correct

    Used with Google Cloud, Direct Peering doesn't produce any custom routes in a VPC network.

  • BDirect Peering reaches internal addresses only after a Cloud Router is attached to the peering sessions

    Incorrect

    Direct Peering connects to Google's edge network; it has no Cloud Router or VLAN attachment in a VPC network.

  • CDirect Peering reaches internal addresses only in regions near the peering facility

    Incorrect

    Location isn't the issue; Direct Peering produces no routes to VPC internal addresses anywhere.

  • DDirect Peering reaches internal addresses, but only with Google's 99.99% SLA purchased separately

    Incorrect

    Google offers no SLA for Direct Peering, and it doesn't reach internal addresses.

Direct Peering exchanges traffic between a business network and Google's edge network to reach Google's public services. It doesn't produce custom routes in a VPC network and has no SLA; reaching VPC internal addresses takes Cloud Interconnect or Cloud VPN.

Question 5 · choose 1

Three service projects use a Shared VPC network whose HA VPN gateway to the data center was created by a network admin in the host project. Finance wants each service project to carry the cost of the outbound traffic that its VMs send to the data center through the VPN. What should the network engineer tell finance?

  1. AThe cost goes to each sending service project, like the rest of its outbound traffic
  2. BThe cost goes to the host project, because that project contains the HA VPN gateway
  3. CThe cost is split evenly among the service projects attached to the host project
  4. DThe cost goes to each service project whose admins hold subnet-level Network User permissions
Show the answer and why
  • AThe cost goes to each sending service project, like the rest of its outbound traffic

    Incorrect

    Outbound traffic through a VPN gateway is the exception: it is attributed to the project that contains the gateway.

  • BThe cost goes to the host project, because that project contains the HA VPN gateway

    Correct

    Outbound traffic through a VPN gateway is attributed to the project that contains it, whichever service project starts the transfer.

  • CThe cost is split evenly among the service projects attached to the host project

    Incorrect

    Shared VPC doesn't split this cost; it goes to the gateway's project.

  • DThe cost goes to each service project whose admins hold subnet-level Network User permissions

    Incorrect

    The permission level decides which subnets an admin may use, not where VPN traffic is billed.

In Shared VPC, the outbound traffic that a service project's VMs generate is normally attributed to that service project. Outbound traffic to VPNs is different: it is attributed to the project that contains the VPN gateway. A gateway created in the Shared VPC network lives in the host project, so the host project is billed whichever service project sends the traffic.

Google Cloud documentation

Question 6 · choose 1

A storage team wants 8,896-byte jumbo frames between its data center and VMs across Dedicated Interconnect. A separate compliance team wants every VLAN attachment to carry only IPsec-encrypted traffic through HA VPN over Cloud Interconnect. What should the network engineer tell both teams?

  1. AJumbo frames work on encrypted attachments once the VPC network's MTU is also set to 8,896 bytes
  2. BJumbo frames work on encrypted attachments if the on-premises router prefragments the packets
  3. CJumbo frames work on any attachment, because IPsec adds no overhead on Cloud Interconnect
  4. DJumbo frames work only on unencrypted VLAN attachments, so the needs can't share one
Show the answer and why
  • AJumbo frames work on encrypted attachments once the VPC network's MTU is also set to 8,896 bytes

    Incorrect

    The network MTU doesn't lift the limit: encrypted attachments for HA VPN over Cloud Interconnect don't support jumbo frames.

  • BJumbo frames work on encrypted attachments if the on-premises router prefragments the packets

    Incorrect

    Prefragmentation handles oversized packets; it doesn't make jumbo frames available on encrypted attachments.

  • CJumbo frames work on any attachment, because IPsec adds no overhead on Cloud Interconnect

    Incorrect

    Encrypted attachments are a documented exception for jumbo frames.

  • DJumbo frames work only on unencrypted VLAN attachments, so the needs can't share one

    Correct

    The 8,896-byte MTU is supported only on unencrypted attachments; HA VPN over Cloud Interconnect doesn't support jumbo frames for encrypted ones.

VLAN attachments support MTUs of 1,440, 1,460, 1,500 and 8,896 bytes, but the 8,896-byte jumbo frame MTU is supported only on unencrypted attachments. The VPN MTU guidance repeats that HA VPN over Cloud Interconnect doesn't support jumbo frames for encrypted VLAN attachments.

Practice domain 1 →Practice all domains →