
Google publishes reference topologies for Cloud Interconnect that qualify for each of its SLAs. The building block in all of them is the edge availability domain. Edge availability domains in the same metro are kept apart for maintenance, so a failure or maintenance event in one domain doesn't take down links in the other. The 99.9% design uses two domains in one metro. The 99.99% design repeats that pattern across two metros and two Google Cloud regions, and it depends on global dynamic routing. The Professional Cloud Network Engineer exam expects you to tell these designs apart and to spot the ones that look redundant but don't qualify.
The 99.9% SLA requires two connections in the same metro, each in a different edge availability domain. Putting both connections in one edge availability domain leaves a single point of failure, because both circuits can fail together during maintenance on that domain, so that layout doesn't qualify. Google positions the 99.9% design for noncritical applications, because it still leaves you exposed to the loss of a whole metro. Two Dedicated Interconnect connections in one metro qualify only for 99.9%, even when they sit in different edge availability domains.
For production and mission-critical workloads, Google recommends the topology that qualifies for the 99.99% SLA. It is also Google's recommended topology for production Partner Interconnect workloads, and it has four parts.
Each part covers a different failure. The split across edge availability domains keeps a metro's traffic flowing when one domain is down. The second metro and region protect against a problem with an entire metro or region. Global dynamic routing ties the two halves together, so traffic still reaches every subnet through the surviving region if a whole metro or region fails. For Professional Cloud Network Engineer scenarios, it helps to check all four parts in turn, because a design that has three of them still falls short.
A design missing any of the four parts reintroduces a single point of failure, whether at an edge availability domain, a metro, a region or a Cloud Router. The routing mode is the part that doesn't show up in the physical layout. With regional dynamic routing, each Cloud Router only exchanges routes for subnets in its own region, so the second region can't take over for the whole network, and the topology doesn't satisfy 99.99% however many connections it has.
Regional resilience can also be designed directly, without building the full 99.99% layout. With global dynamic routing, a Cloud Router applies the on-premises routes it learns to every region in the VPC network and advertises subnets from all regions to the on-premises router, so one Cloud Router can carry hybrid traffic for the whole network. To survive the loss of a whole region, the hybrid path has to exist in at least two different regions.
Redundancy inside a single region is a different kind of protection. Extra attachments and Cloud Routers in the same region guard against one of them failing, but they share the region's fate, so they don't keep hybrid connectivity up if the region itself is lost.
Our Professional Cloud Network Engineer practice exams cover Cloud Interconnect SLA topologies alongside VLAN attachment placement and regional versus global dynamic routing, with detailed explanations that drill these distinctions.