Prepare and test your skills

Prepare and test your skills

Creating a VPC with subnets is the practice of building a global private network and dividing it into regional address blocks. A Virtual Private Cloud (VPC) network carries internal traffic between resources, while each subnetwork holds IP addresses for one region. The network mode and address plan decide how much control you keep and whether future connections will work.
Auto mode and custom mode VPC networks are the two ways Google Cloud creates subnetworks. An auto mode network automatically creates one subnetwork in every region with a predefined IPv4 range, and it gains a new subnetwork whenever Google opens a new region. A custom mode network starts with no subnetworks, so you create each subnetwork manually by choosing its region and IPv4 range. All subnetworks in the same VPC can still reach each other because the VPC is one global network.
| Network mode | How subnetworks are created | When to use it |
|---|---|---|
| Auto mode | Google creates one subnetwork per region automatically | Small projects, labs, and quick starts where ease matters more than control |
| Custom mode | You create each subnetwork and choose its region and IP range | Production and connected environments where ranges must be planned to avoid overlap |
Exam tip: An auto mode VPC network can be converted to custom mode, but a custom mode network cannot be converted back to auto mode.
Regional subnet planning is the process of assigning a non-overlapping Classless Inter-Domain Routing (CIDR) range to each region. Each regional subnet is tied to one region, such as 192.168.255.0/24, and resources in that region receive addresses from that block. Non-overlapping ranges prevent routing conflicts with the same VPC, peered VPC networks, and on-premises networks connected by Virtual Private Network (VPN) or Interconnect. Good planning also reserves space for future regions, test and production separation, and private service connections.
Three practical checks help avoid overlap before creating a subnet:
gcloud compute networks subnets list and choose a range that does not appear.Subnet sizing also affects availability because every subnet reserves 4 addresses for network functions. Usable addresses equal total addresses minus 4, so a /29 block with 8 total addresses leaves only 4 addresses for nodes. A Cloud Run service with direct VPC access cannot start new instances when its subnet runs out of IPv4 addresses, and expanding a private service allocation must only increase the range, for example from 10.0.10.0/24 to at least 10.0.10.0/23.
A Shared VPC architecture is a central VPC network that lives in one host project and serves workloads in other service projects. The host project contains the Shared VPC network, its subnets, firewall rules, and VPN connections. A service project attaches to the host project so its instances can use the host subnets, while a standalone project uses neither role. A Shared VPC Admin enables the host project, attaches service projects, and grants access, while a Service Project Admin manages instances inside one service project.
Setting up a Shared VPC follows a fixed order. First the Shared VPC Admin designates a project as the host project. Then the Shared VPC Admin attaches one or more service projects to that host project. Then the Shared VPC Admin grants the Network User role (compute.networkUser) for the whole host project or for selected subnets. Attaching a service project alone grants no subnet access until this role is granted.
| Sharing scope | Scope of access | Choose when |
|---|---|---|
| Project-level permissions | All current and future subnets in the host project | Every service project should be able to use any subnet |
| Subnet-level permissions | Only selected subnets in the host project | Different teams should be limited to specific subnets or regions |
Isolating environments with Shared VPC means using separate host projects for test and production. One host project can hold a testing network used only by testing service projects, while another host project holds a production network used only by production service projects. Both networks can safely reuse the same CIDR ranges, such as 10.0.1.0/24 in us-west1, because each instance belongs to a different Shared VPC network. Instance creation still splits ownership: the subnet belongs to the host project, the instance belongs to the service project, the instance receives a free address from the shared subnet, and the instance must be in a zone in the same region as that subnet.
A hybrid Shared VPC connects one host network to on-premises through Cloud VPN. Traffic flows from the on-premises network through the VPN connection into the host Shared VPC network, and then to instances in attached service projects. Separate teams can manage each service project with no permissions in the others, while the Shared VPC Admin controls whether each team receives one subnet, another subnet, or project-level access to any subnet.
Continue with this study guide: Creating a VPC with subnets (e.g., custom mode VPC, Shared VPC)
Creating and applying Cloud Next Generation Firewall (Cloud NGFW) (opens in a new tab) policies is the practice of grouping firewall rules and attaching them to the right networks. A firewall policy (opens in a new tab) is a container that groups multiple rules, while a firewall policy rule is the fundamental unit that allows, denies, or inspects traffic. The policy type sets the scope, and association activates the policy on a specific resource.
Firewall policy scope determines which VPC networks a set of rules can protect. Hierarchical firewall policies are created at the organization or folder level and must be associated with an organization or folder, so their rules flow down to every matching VPC network below. Global network firewall policies live in the same project as the VPC networks they protect and apply to all regions of each attached VPC network. Regional network firewall policies also live in the same project but apply to a single region of one or more VPC networks.
| Policy type | Scope of protection | Association target |
|---|---|---|
| Hierarchical firewall policies | Many VPC networks in one or more projects | Organization or folder |
| Global network firewall policies | All regions of attached VPC networks | One or more VPC networks in the same project |
| Regional network firewall policies | Single region of attached VPC networks | Single region of one or more VPC networks in the same project |
Google Cloud also creates regional system firewall policies when a Google service requires rules in a region. Those system rules cannot be modified except to enable or disable logging, carry no charge for evaluation, and are evaluated immediately after hierarchical policies. A policy has no effect until association is complete, and Remote Direct Memory Access (RDMA) networks are protected by regional network firewall policies.
Rule direction specifies whether a firewall policy rule (opens in a new tab) controls inbound or outbound traffic from the perspective of the target. Every rule applies to either ingress or egress, never both at the same time. Ingress rules protect targets such as virtual machine (VM) network interfaces and managed Envoy proxies, while egress rules control what those targets can send outward.
| Rule direction | Traffic flow | Source requirement | Destination requirement |
|---|---|---|---|
| Ingress | Inbound packets arriving at the target from a source | Mandatory | Optional |
| Egress | Outbound packets sent from the target to a destination | Optional | Mandatory |
An ingress rule with a deny action blocks incoming connections to its targets, while an egress rule with an allow action permits a VM to send traffic to listed destinations. Source describes the sender of inbound traffic, and destination describes the receiver of outbound traffic. Protocols and destination ports further narrow the match, and the action decides the outcome: allow the traffic, deny the traffic, or send matching traffic to inspection for deeper checks.
Targets, priority, and stateful filtering together decide which resources receive a rule, in what order rules are checked, and how return traffic is handled. Target type identifies the kind of resource the rule protects, which is VM instances by default or managed Envoy proxies used by internal Application Load Balancers and internal proxy Network Load Balancers. Target narrows the rule to specific instances or proxies so one policy can protect different workloads differently. Priority is a unique integer within a policy where a lower number means higher priority and is evaluated first, so the first matching rule controls the connection even if a lower-priority rule would decide differently.
Stateful filtering regulates traffic at Layer 3 and Layer 4 of the OSI model without re-checking every packet. The firewall remembers active connections in a connection tracking table once a rule allows them. Return packets for an allowed connection are then permitted automatically, which protects workloads from external and internal threats while allowing legitimate two-way communication.
Exam tip: Source is mandatory for ingress and optional for egress, while destination is mandatory for egress and optional for ingress; a lower priority number means the rule is evaluated first.
Kubernetes Network Policies control pod-to-pod traffic at the IP address or port level, which corresponds to OSI Layer 3 or Layer 4. Policies are application-centric, so they specify how a pod may communicate with other pods in the same cluster. If no pod-to-pod restrictions are needed, no policies are required. When isolation is needed, for example for Cassandra and Redis data pods, the policy allows only Runtime, Synchronizer, and Mart pods to reach them while blocking Ingress, Watcher, and other unrelated pods. Enforcement requires a running cluster with policy enforcement enabled through the platform vendor, and logging examples (opens in a new tab) can help verify how firewall decisions are recorded.
Using tags and service accounts in Cloud NGFW policy rules is a way to enforce firewall decisions by workload identity instead of changing IP addresses. A service account is an identity attached to a compute instance, while secure tags are centrally governed labels bound to instances. Both let rules follow the workload as addresses change.
Central control with hierarchical policies means defining firewall rules once at a high level and enforcing them across many projects. A hierarchical security policy is created and owned at the organization or folder level, but its rules apply to no resources at creation time. Enforcement begins only after association with an organization, folder, or project, and child projects then inherit the associated rules automatically. This flow reduces misconfigurations because a central block of a malicious IP range, a geography restriction, a critical vulnerability mitigation, or a compliance rule reaches every associated child project without separate updates.
Hierarchical and service-level policies differ in attachment and actions. A hierarchical policy attaches to organization, folder, or project nodes and supports Allow, Deny, Redirect with only EXTERNAL_302, and Goto next, where Goto next passes evaluation to the next rule for layered rule sets. A service-level policy attaches directly to a backend service or backend bucket for a global external or Classic Application Load Balancer and also supports throttling and reCAPTCHA redirects for that single service.
Service account-based filtering controls VPC traffic by naming the identities on both sides of a connection. When a VM is created to run as a specific service account, all traffic from that VM carries that identity for firewall evaluation. A Cloud NGFW rule names a source service account for the VM that initiates traffic and a destination service account for the VM that receives it. Traffic flows from the source instance to the enforcement point, where the rule compares both identities and then allows or denies the connection. The rule applies only when both identities match, so other VMs in the same VPC or subnet are unaffected.
| Filtering method | What the rule matches | Behavior when IPs change |
|---|---|---|
| IP-range filtering | Source or destination IP ranges, such as a subnet CIDR | The rule must be updated when addresses change or new instances are added |
| Service-account filtering | Source and destination service accounts attached to instances | The rule continues to work because the identity stays with the workload |
Filtering without static ranges improves security because administrators do not need to predict every address or rewrite rules during scale-up, scale-down, or replacement. Narrow rules that name only the required source and destination service accounts enforce least privilege, for example allowing a frontend service account to reach a backend service account without opening broad IP ranges.
Secure tags are Identity and Access Management (IAM)-governed Resource Manager tags that let Cloud NGFW enforce rules by workload identity. A secure tag key used with Cloud NGFW must have its purpose set to GCE_FIREWALL and must include a purpose-data attribute that cannot be changed after creation. The parent can be a project or an organization, and each key can have up to 1,000 unique tag values created by principals with the Tag Administrator role. Purpose-data controls scope: a key can specify a single VPC network in the same project or organization, or it can specify organization=auto for all VPC networks in the organization.
To use a secure tag, a tag value is bound to a VM instance, and the firewall rule then applies to the network interfaces that the bound value covers. For each tag key, only one of its values can be bound to a given instance, but the same value can be bound to unlimited instances. Google Cloud enforces a limit of 10 tag values that apply to each network interface and blocks additional bindings that would exceed it. A value scoped to one network applies only to interfaces in that network, while a value scoped to organization=auto can apply across the organization. Secure tags can also identify interfaces in peered networks when scope and connectivity allow, and they can be bound to Google Kubernetes Engine (GKE) clusters and node pools.
Packet matching with secure tags depends on direction and role. A source secure tag in an ingress rule matches packets sent from a source interface that carries the value. A target secure tag in an ingress rule matches packets received by a target interface that carries the value. A target secure tag in an egress rule matches packets sent from a target interface that carries the value. Secure tag values are used as sources for ingress rules and as targets for ingress or egress rules in hierarchical, global network, and regional network firewall policies.
Secure tags compared with network tags show the difference between governed identity and simple labels. Network tags are simple character strings added to VM instances or templates without access controls, so anyone who can edit a VM can add or remove them. Secure tags are IAM-governed objects with a defined parent, purpose, and scope, so only authorized administrators and tag users can define and assign them. This separation supports micro-segmentation, where a network is divided into small zones with granular rules for east-west internal traffic and north-south external traffic.
| Capability | Secure tags | Network tags |
|---|---|---|
| Access control | Governed by IAM | No access control, simple string on VM or template |
| Definition | Tag key with up to 1,000 values in a project or organization | Simple string in a project |
| Supported policies | Hierarchical, global network, and regional network firewall policies | VPC firewall rules only |
Exam tip: Use secure tag values in hierarchical, global, and regional network firewall policies; use network tags only in VPC firewall rules.
Establishing network connectivity is the choice of how private networks exchange traffic inside Google Cloud and with external environments. VPC Network Peering connects two VPC networks with internal addresses, Cloud VPN connects over the encrypted public internet, and Cloud Interconnect connects over a private physical circuit. The right choice depends on privacy, throughput, latency, and management effort.
VPC Network Peering connects two separate VPC networks so resources communicate using private, internal IP addresses. Traffic between peered networks stays internal and does not travel over the public internet. Each network continues to manage its own subnets, firewall rules, and routes. A VM in one network reaches a VM in the peered network by its primary internal IP address from its subnet primary CIDR range, and it can also use optional alias IP ranges for extra addresses on the same interface.
Managing non-overlapping ranges is required because each internal address must route to only one destination. In custom-mode networks every subnet needs one primary CIDR range, with secondary ranges optional, and an alias range must come from the primary or secondary range of the same subnet resource. An alias can be a full CIDR such as 10.128.1.0/24, a single address such as 10.128.7.29/32, or an auto-allocated range by netmask such as /24. Google Cloud configures internal DNS only for the primary IP of the primary interface, so alias addresses and secondary interfaces need manual DNS when names must work across peered networks. All ingress and egress traffic, including traffic to alias ranges, is still checked by VPC firewall rules.
Peering is non-transitive, which limits how connectivity spreads. If network A is peered with network B, and network B is peered with network C, network A does not automatically gain connectivity to network C. Each pair that needs direct communication requires its own peering. Routing has a related boundary: a static route with a next-hop instance must use a primary internal IPv4 address on a VM interface in the route own VPC network, not an alias IP or an address in a peered network.
Exam tip: A static route next hop must be a primary internal IP in the route's own VPC network; an alias IP or an address in a peered VPC network cannot be used as the next hop.
Cloud VPN connects a VPC network to an external environment over the public internet using secure, encrypted tunnels. Traffic flows from the on-premises network to the VPC through an IPsec tunnel, and return traffic flows back through the same encrypted path. Border Gateway Protocol (BGP) provides dynamic routing by automatically exchanging routing and reachability information between the two sides. Each autonomous system using BGP is identified by a unique autonomous system number (ASN), and peers can use MD5 authentication where both sides share the same key before the connection is established.
High Availability (HA) VPN is the deployment that provides encrypted internet connectivity with high reliability. HA VPN scales through Equal-Cost Multi-Path (ECMP) routing across multiple tunnels and carries a 99.99% service availability SLA. Gateway maximum transmission unit (MTU) is the largest IP packet including IPsec overhead supported by a tunnel, while payload MTU is the largest original packet without IPsec overhead that can be sent without fragmentation. Because encapsulation adds overhead, payload MTU is lower than gateway MTU.
Cloud VPN fits when private networks must be joined over the internet with encryption and dynamic routing:
Its tradeoffs center on management, cost, and throughput:
Cloud Interconnect creates a direct physical connection between an on-premises network and a Google Cloud VPC network. Traffic flows from the on-premises router across a private circuit to the Google network edge and then into the VPC, without traveling over the public internet. This private path provides lower latency, higher bandwidth, and more consistent performance than internet-based VPN, which makes it the choice for high-volume and reliability-sensitive hybrid workloads.
| Option | How it connects | Choose when |
|---|---|---|
| Dedicated Interconnect | Direct physical circuit from customer routers to Google in a colocation facility | Data volume is very high, reliability needs are strict, and the team can deploy in a colocation facility |
| Partner Interconnect | Circuit through a supported partner provider to Google | Direct facility access is not available, startup must be faster, or bandwidth needs are lower and variable |
Both options keep hybrid traffic off the public internet and depend on Cloud Router to exchange routes with BGP. First the VLAN attachment connects the circuit to the VPC, then Cloud Router learns on-premises routes and advertises VPC routes back. Traffic can flow in both directions only after the attachment exists and BGP sessions are established. Dedicated Interconnect suits large migrations and steady replication because the organization controls the full circuit, while Partner Interconnect suits flexible needs because the partner manages the physical link to Google.
Choosing and deploying load balancers is the process of placing application backends behind a single address that distributes client traffic. Cloud Load Balancing (opens in a new tab) provides external addresses for internet clients and internal addresses for VPC traffic. The choice depends on client location, protocol, backend scope, and failover needs.
Global load balancing serves applications whose clients arrive from different parts of the world. Backends are created in the region closest to where traffic arrives at the Google frontend, and in many cases backends run in multiple regions to minimize latency. The closer users are to the backends, the lower the network latency, so multi-region placement delivers more consistent performance. The cross-region internal Application Load Balancer is a proxy-based Layer 7 balancer based on Envoy that balances traffic to globally distributed backend services. Traffic flows from the forwarding rule to the closest backend relative to the forwarding rule, Virtual IP addresses (VIPs) are allocated from a subnet in a specific region, and VIPs from multiple regions can share the same global backend service. If one region backends are down, traffic can fail over to another region for high availability.
Protocol choice affects global performance. For the global external Application Load Balancer and classic Application Load Balancer, HTTP/3 built on IETF QUIC is recommended. HTTP/3 is enabled by default in major browsers, Android Cronet, and iOS, clients without support are not impacted, but UDP must not be blocked or rate-limited and HTTP/3 must not be disabled on the load balancers. Turning on Cloud CDN with a default global external Application Load Balancer can further reduce latency by serving cacheable responses closer to clients.
Regional load balancing keeps all backends in one Google Cloud region for applications with localized traffic. The regional internal Application Load Balancer is also an Envoy-based Layer 7 balancer, but it requires backends in a single region. After configuration it automatically allocates Envoy proxies to meet traffic needs and applies traffic control based on HTTP or HTTPS parameters. The regional external Application Load Balancer follows the same localized pattern for internet-facing HTTP and HTTPS traffic. For regional internal mode, clients can be limited to the same region or can be in any region, based on whether global access is disabled or enabled on the forwarding rule. Choose regional balancing when compliance, data locality, or application design keeps workloads in one place and cross-region failover is not required.
| Mode | Backends | Client access | High availability |
|---|---|---|---|
| Cross-region internal Application Load Balancer | Globally distributed services, traffic to closest backend relative to forwarding rule | VIP from a subnet in a specific region, VIPs from multiple regions can share one global backend service | Traffic fails over to another region if one region backends are down |
| Regional internal Application Load Balancer | Backends must be in a single region | Clients limited to that region or in any region, based on global access setting | Envoy proxies auto-allocated to meet traffic needs within the region |
Layer 7 and Layer 4 handling match client traffic to the OSI layer where routing decisions are made. Application Load Balancer (HTTP/HTTPS) (opens in a new tab) works at Layer 7, terminates client HTTP and HTTPS connections, and routes by host name and request path. Layer 4 proxy and passthrough services handle raw TCP and User Datagram Protocol (UDP) without reading HTTP content and route only by IP address, port, and connection information.
The Layer 7 proxy uses a URL map with host rules and path matchers. A host rule directs requests by host name such as example.com, while a path matcher directs by request path such as /images versus /api. A simple setup can keep the default host and path so all requests go to one backend service, while a multi-service setup adds rules to split traffic. Layer 4 backends can use client IP session affinity, where the balancer hashes the client IP and the forwarding-rule IP to keep a client on the same backend VM for stateful TCP services.
| Traffic type | OSI layer | Routing basis | Choose when |
|---|---|---|---|
| Application Load Balancer (HTTP/HTTPS) | Layer 7 | Host name, request path, URL map rules | Traffic is HTTP/HTTPS and needs content-based routing, TLS termination, or serverless backends |
| Proxy Network Load Balancer for TCP | Layer 4 | IP address, port, session affinity | Traffic is TCP/UDP, TLS should stay encrypted to the backend, or HTTP inspection is not needed |
Regional Envoy-based balancers depend on a proxy-only subnet with purpose Regional Managed Proxy, active in the balancer region, commonly using a range such as 10.129.0.0/23 alongside a main subnet such as 10.1.2.0/24. Traffic from the balancer to serverless network endpoint group (NEG) backends uses special routes outside the VPC that are not subject to firewall rules, so a balancer with only serverless NEG backends does not need firewall rules from the proxy-only subnet.
Exam tip: Choose Application Load Balancer when routing depends on host or path in the URL; choose Layer 4 TCP/UDP handling when routing depends only on IP address and port.
Backend services, health checks, and frontend configuration connect the public entry point to healthy backends. A backend service groups backends such as managed instance groups and links them to a health check and a routing rule. Traffic flows from the URL map to the backend service, then to each attached backend in different zones. A typical backend setup follows this order:
Health checks are probes that test whether application software on each backend is actually responsive. A Compute Engine VM can remain in RUNNING state while its application has crashed, so autoscaling alone cannot guarantee usable capacity. The health check is attached when the backend service is created, and the balancer sends traffic only to backends that answer correctly. When a backend stops responding, the balancer stops sending it new traffic, which improves reliability and prevents healthy resources from becoming overloaded.
The frontend provides the stable entry point through forwarding rules, target proxies, and certificates. Traffic flows from clients to a forwarding rule matched on IP, protocol, and port, then to a target HTTP, HTTPS, or TCP proxy, then to the URL map. The chain separates connection handling, routing logic, and backend selection:
us-central1.| Frontend | Protocol | Port | Certificate requirement |
|---|---|---|---|
| HTTP frontend | HTTP | 80 | No certificate required |
| HTTPS frontend | HTTPS | 443 | Requires a certificate map or classic Compute Engine SSL certificate |
Both frontends commonly use Premium Network Service Tier, IPv4, a reserved static IP, and an HTTP keepalive timeout of 5 to 1200 seconds with a default of 610 seconds. Google-managed certificates are recommended for production because Google obtains and renews them automatically but require domain ownership, while self-managed and self-signed certificates suit brought keys and testing. Enabling HTTP to HTTPS redirect creates an additional partial HTTP balancer that uses the same IP and redirects HTTP to HTTPS, which requires HTTPS protocol plus a reserved IP.
Continue with this study guide: Choosing and deploying load balancers
Differentiating Network Service Tiers is the choice of which network carries traffic between the internet and Google Cloud. Network Service Tiers (opens in a new tab) controls this path for each resource. Premium Tier uses Google global backbone for performance, while Standard Tier uses the public internet for lower cost.
Premium Tier routing carries traffic on Google private fiber network with over 200 points of presence (PoPs) for low-latency delivery. Inbound traffic from an internet user enters Google network at the PoP closest to that user and travels inside Google network to the VM or Cloud Storage bucket. Outbound traffic stays on Google network until it exits at the PoP closest to its destination, which reduces hops and congestion. Because most of the journey stays inside the managed network, latency is lower and more consistent.
Premium Tier supports both regional and global external IP addresses, and all global external IP addresses must use Premium Tier. Applications with backends in more than one region, such as external Application Load Balancers and external proxy Network Load Balancers spanning regions, require Premium Tier. Premium Tier offers a 99.99% uptime service level agreement (SLA) for VMs, where a VM that loses external connectivity for more than a minute counts as downtime.
Standard Tier routing carries traffic over public internet transit for regional distribution. Inbound traffic enters Google network at a peering or transit PoP near the region of the Standard Tier IP address, not near the user. Outbound traffic normally leaves Google network from the same region as the sending VM or bucket, regardless of destination, and is then carried by transit providers and ISPs. Because performance depends on public routing and congestion outside Google control, latency is typically higher and more variable.
Standard Tier is priced lower than Premium Tier and fits applications that are not sensitive to latency or that keep VMs and storage in a single region. Standard Tier includes a Free Tier of 200 GB of free usage per month in each region used, measured per SKU across all projects. Standard Tier offers a 99.9% uptime SLA for VMs, which is lower than Premium Tier, so the saving comes with reduced performance and reliability guarantees.
| Tier | Routing path | Performance characteristic |
|---|---|---|
| Premium Tier | Traffic travels from the hosting region across Google global private fiber and exits close to the user | Low-latency, consistent global delivery |
| Standard Tier | Traffic exits Google network in the hosting region and travels over public internet transit | Higher, more variable latency focused on regional delivery |
Exam tip: Premium Tier exits Google's network close to the user, while Standard Tier exits Google's network in the same region that hosts the resource.
Tier selection balances performance needs, geography, and pricing, and the tier is chosen at the resource level so different resources can use different tiers. If you are not sure which tier fits, choose Premium Tier because it is the default. Use Premium Tier for global users and multi-region backends, and use Standard Tier for single-region, cost-sensitive workloads. Standard Tier with Cloud Storage is allowed only when the bucket is a backend of an external Application Load Balancer in the same region, while multi-regional buckets as backends require Premium Tier.
The network tier can be set as a project default and overridden for individual resources, and the resource-level setting always wins. The project default applies only to newly created eligible resources and does not change existing resources. For a VM or instance template the tier is set on its external IP address, and for load balancing the tier is set on the forwarding rule and its external IP address. Classic Application Load Balancers and classic proxy Network Load Balancers in Standard Tier must meet three related requirements:
With Standard Tier, clients from anywhere can still reach the external address, but a Google Front End terminates the connection in the chosen region and contacts backends only in that region. To serve multiple regions with Standard Tier, you must create a separate regional balancer with its own regional IP in each region. External addresses come from separate pools by scope and tier, so an address cannot move between pools. Changing an ephemeral IP from one tier to the other changes its IP address, and changing a balancer tier requires deleting the forwarding rule and reserving a new IP of the new tier. Unused Standard Tier addresses are automatically parked, meaning HTTP requests to them can receive a Google-hosted HTTP 404 page, and a Standard Tier address on a VM can appear parked during a live migration longer than 20 minutes.
Continue with this study guide: Differentiating Network Service Tiers
An auto mode Virtual Private Cloud network automatically creates one subnetwork in every region with a predefined IPv4 range and gains a new subnetwork whenever Google opens a new region, while a custom mode network starts with no subnetworks and requires manual creation of each subnetwork by region and IPv4 range. An auto mode network can be converted to custom mode, but a custom mode network cannot be converted back to auto mode.
In a Shared VPC architecture the Shared VPC network, its subnets, firewall rules and VPN connections live in one host project and service projects attach to that host project to use its subnets. Attaching alone grants no subnet access until the Shared VPC Admin grants the Network User role for the whole host project or for selected subnets, and each instance belongs to its service project while using a free address from the shared subnet in the same region as that subnet.
Every Cloud Next Generation Firewall policy rule applies to either ingress or egress, never both, where ingress controls inbound packets arriving at the target and egress controls outbound packets sent from the target. Source is mandatory for ingress and optional for egress while destination is mandatory for egress and optional for ingress, and a lower priority number means higher priority and is evaluated first.
Premium Tier carries traffic on the Google private fiber network with over 200 points of presence for low-latency global delivery and offers a 99.99 percent uptime service level agreement for virtual machines, while Standard Tier routes traffic over public internet transit near the resource region with higher and more variable latency and offers a 99.9 percent uptime service level agreement. Use Premium Tier for global users and multi-region backends and use Standard Tier for single-region cost-sensitive workloads, and choose Premium Tier when uncertain because it is the default.
Last updated