Planning Network Resources

Google ACE ACE · Planning and implementing a cloud solution

By Abu Anns · Last updated 20 November 2026Exam version: ACE (exam guide live 30 June 2026)

What network planning means for the exam

Network planning on the ACE exam is about making three choices before you build anything: how traffic will be distributed (load balancing), which network quality and cost trade-off you want (Network Service Tiers), and where resources will live (regions and zones). These are planning decisions because they are hard or impossible to change later — a load balancer's type, a bucket's location, and a subnet's region are all fixed at creation. The exam tests whether you can make these choices from a workload description, not whether you can recite product names.

Load balancing: the decision tree

Google Cloud's load balancers fall into two families: global load balancers, which use a single anycast IP address to route users to the nearest healthy backend anywhere in the world, and regional load balancers, which distribute traffic within one region. Within each family there are Layer 7 (application) load balancers that understand HTTP(S) and can route by URL path or host, and Layer 4 (network) load balancers that route TCP/UDP traffic without inspecting the application layer. The decision tree the exam expects is roughly this:

NeedLoad balancer
Global HTTP(S) traffic, single IP, content-based routingGlobal external Application Load Balancer
Regional HTTP(S) traffic, or compliance requiring in-region terminationRegional external Application Load Balancer
HTTP(S) traffic inside the VPC onlyInternal Application Load Balancer
Global TCP/UDP traffic (non-HTTP)Global external Network Load Balancer (passthrough or proxy)
TCP/UDP traffic inside the VPC onlyInternal Network Load Balancer (passthrough)

The key distinctions to internalize: if the traffic is HTTP(S) and the audience is global, the global external Application Load Balancer is almost always the answer — it gives one anycast IP, terminates TLS, and routes users to the closest healthy region. If the traffic must stay inside the VPC (for example, a backend service that only front-end VMs should reach), an internal load balancer is the answer, because it has no public IP at all. If the traffic is not HTTP — a game server, a database protocol, raw TCP — a Network Load Balancer is the answer, because an Application Load Balancer only handles HTTP(S). The exam loves the trap of offering an Application Load Balancer for non-HTTP traffic.

Two supporting concepts appear regularly. Health checks determine whether a backend is healthy enough to receive traffic; a load balancer stops sending traffic to unhealthy backends automatically. Backend services group the backends (instance groups, network endpoint groups, or serverless backends such as Cloud Run) and hold the configuration: the health check, the session affinity setting, and the balancing mode. A managed instance group is the usual backend for VM-based workloads, and Cloud CDN can be enabled on an Application Load Balancer to cache static content at Google's edge locations worldwide.

Network Service Tiers

Network Service Tiers let you choose the quality of the network path for traffic leaving Google Cloud. Premium Tier (the default) routes egress traffic over Google's private global backbone for as much of the journey as possible, exiting to the public internet close to the user — lower latency and higher reliability, at a higher price. Standard Tier hands traffic off to the public internet at the Google edge location nearest the resource, so the rest of the journey crosses the public internet — cheaper, but with higher and more variable latency. The tier is chosen per resource (a VM's external IP or a load balancer's forwarding rule), and it only affects egress traffic; ingress is unaffected. The exam pattern is simple: when the requirement emphasizes performance and low latency for a global audience, Premium Tier; when it emphasizes cost and the audience is tolerant of latency, Standard Tier. Note that global load balancing requires Premium Tier — a global external load balancer cannot be created on Standard Tier.

Regions, zones, and resource locations

Google Cloud's geography has three levels: regions (a geographic area, such as europe-west2 or us-central1), zones (isolated locations within a region, such as europe-west2-a), and multi-regions (for some storage services). Most resources are zonal (a VM, a zonal disk) or regional (a subnet, a regional managed instance group, a regional disk); a few are global (VPC networks, global load balancers, firewall rules). Placement decisions matter for three reasons: latency (put resources near their users), availability (spread across zones so a zone failure does not take everything down), and cost and compliance (prices vary slightly by region, and some data must stay in a particular geography).

The availability pattern the exam tests most is the regional managed instance group: identical VMs spread across two or three zones in a region, behind a load balancer, so that losing one zone still leaves the service running. A zonal deployment is cheaper and simpler but has a single point of failure. For data, the same idea appears as regional disks (synchronous replication across two zones) and multi-region buckets. When a question says "must remain available if a zone fails", the answer involves spreading across zones — regional MIGs, regional disks, or multi-region storage — and never a single-zone deployment.

Bringing it together

A typical well-architected answer for a global web application looks like this: a global external Application Load Balancer with Cloud CDN on Premium Tier, backend services pointing to regional managed instance groups in two or more regions, with health checks removing unhealthy VMs. That single sentence covers the load balancer choice, the tier choice, and the location choice — and the exam often asks for exactly one of those three decisions at a time, with the others given in the scenario. Read the scenario for the constraint that decides the question: global audience, internal-only traffic, non-HTTP protocol, cost sensitivity, or zone-failure tolerance.

A worked example: planning an e-commerce launch

Consider a company launching an online store for customers in Europe and North America, with a web front end, an API, and a database. The planning decisions fall out of the requirements one by one. The front end is public HTTP(S) traffic with a global audience, so the global external Application Load Balancer with Cloud CDN is the choice, on Premium Tier for the lowest latency. The API is also HTTP(S), but it is only called by the front end — so it sits behind an internal Application Load Balancer, with no public IP at all. The database is not load-balanced (databases use their own replication), but it needs a regional deployment for availability. The VMs run in regional managed instance groups in two regions, one per continent, so a zone failure — or even a whole region failure — leaves the store running. Each decision follows from one constraint in the scenario, and that is exactly how exam questions are built: one requirement, one decision, one correct answer.

Notice what the example does not include: there is no Network Load Balancer, because all the traffic is HTTP(S); there is no Standard Tier, because the audience is global and latency-sensitive; and there is no single-zone deployment anywhere, because availability is a stated requirement. When you read an exam scenario, try the same elimination: cross out the options that contradict a stated requirement, and the answer is usually the only one left standing. The most common planning mistake is choosing the most powerful or most expensive option when the scenario asks for the cheapest adequate one — or the reverse, choosing the cheapest option when the scenario demands availability or performance. Match the choice to the constraint, not to your instinct about what is "best".

Advertisement

Exam traps to watch for

Practice questions

Test yourself on this guide. Answers are checked instantly, with a full explanation for each question.

Q1.A company is launching a public website for users worldwide. They want a single IP address, TLS termination, and users routed to the nearest healthy region. Which load balancer should they choose?

Report an error

Q2.A backend database service should only be reachable from application VMs inside the same VPC. No internet access to it is required. Which load balancer should front it?

Report an error

Q3.A multiplayer game server uses a custom UDP protocol and serves players globally. Which load balancer is appropriate?

Report an error

Q4.A company wants the lowest possible latency for its global users and is willing to pay more for network performance. Which Network Service Tier should it choose for its VMs' external IPs?

Report an error

Q5.A startup wants to minimize network costs for a batch job that uploads results to users once a day. Latency is not important. Which Network Service Tier is the best fit?

Report an error

Q6.An application must remain available if an entire zone fails. Its VMs are currently all in europe-west2-a. What should you do?

Report an error

Q7.Which statement about load balancer health checks is correct?

Report an error

Q8.A team is planning a global external Application Load Balancer. What must they know about Network Service Tiers for this design?

Report an error