Prepare and test your skills

Prepare and test your skills

2.1 Planning and implementing compute resources is resource planning (opens in a new tab) for matching each workload to the right compute service, virtual machine setting, and scaling model. Compute Engine provides virtual machines with full operating system control, Google Kubernetes Engine (GKE) provides managed Kubernetes orchestration, and Cloud Run provides serverless containers with automatic scaling. The same container image in Artifact Registry can run on GKE and Cloud Run without modification.
Choosing compute means trading control for simplicity across Compute Engine, GKE, and Cloud Run. Choose Compute Engine (opens in a new tab) when the workload needs full administrator access, custom kernels, or specialized GPUs and TPUs. Choose GKE when complex microservices need Kubernetes orchestration, custom networking, or persistent storage, while choose Cloud Run when stateless, request-driven services need minimal operations. Google manages the GKE control plane for security and availability, while the user manages nodes, networking, scaling, and upgrades.
GKE supports stateless and stateful applications and attaches persistent volumes so data survives container restarts. In Autopilot mode Google provisions and scales nodes automatically and the user pays only for resources requested by workloads, while in Standard mode the user manages node pools directly. Capacity adjusts with Horizontal and Vertical Pod Autoscalers, which add or remove containers, and the cluster autoscaler, which adds or removes nodes. Cloud Run runs services that respond to web requests and jobs that run background tasks and then exit, with scale to zero when idle and pay-per-use billing rounded up to the nearest 100 milliseconds.
| Option | Management | Choose when |
|---|---|---|
| Compute Engine | User manages OS, middleware, and apps | Full OS access or specialized hardware is required |
| GKE | Google manages control plane; user manages nodes and networking | Stateful apps or complex microservices need orchestration |
| Cloud Run | Google manages OS, network, and capacity | Stateless, request-driven services need minimal operations |
Launching a Compute Engine instance means selecting machine type, boot image, availability behavior, and secure access at creation time. The machine type sets virtual CPUs and memory, with predefined types such as e2-small or custom types with specific CPU and memory. The boot disk holds the operating system, where a public image is supplied by Google and a custom image is created from a configured disk for a standardized environment. An instance template captures machine type, image, network, and disks as a reusable blueprint.
Availability depends on provisioning model and placement across zones. Standard capacity is regular capacity, Spot is reclaimable spare capacity, and reservations ensure capacity in a zone. A workload in a single zone stops serving if that zone fails, so protection requires spreading VMs across zones and distributing traffic with Cloud Load Balancing. Traffic flows from the load balancer to healthy backend instances, while an autoscaler adds or removes instances as demand changes.
Block storage choice controls persistence, replication, and performance. Zonal Persistent Disk keeps one copy in one zone and the VM must run in that same zone, so recovery from zone loss depends on snapshots. Regional Persistent Disk keeps two synchronously replicated copies in two zones of the same region, so a VM in the surviving zone can continue without a restore. Google Cloud Hyperdisk (Hyperdisk) lets IOPS, throughput, and capacity scale independently, with Balanced for general use, Extreme for up to 350,000 IOPS, Throughput and ML for sequential throughput, and Balanced High Availability for synchronous replication across two zones.
An autoscaled managed instance group (MIG) is a collection of identical VMs managed as one entity from an instance template. Only managed groups support autoscaling; unmanaged groups contain heterogeneous VMs managed manually. A template cannot be updated after creation, so changes require a new template and an update to the MIG. A custom image with preinstalled apps starts faster, while a public image with a startup script is more flexible.
Autoscaling requires at least one scaling signal plus minimum and maximum instance counts, and when multiple signals exist the autoscaler applies the largest recommendation. Signals include average CPU utilization toward a target such as 0.60, load-balancing serving capacity, Cloud Monitoring metrics, schedules, and Pub/Sub queue depth for zonal MIGs. The initialization period defaults to 60 seconds to protect scale-out from incomplete startup data, and the stabilization period defaults to 600 seconds to verify lower load before removing VMs. A zonal MIG defaults to 1,000 VMs in one zone, while a regional MIG defaults to 2,000 VMs across zones for survival if one zone fails.
Spot VM instances run on excess capacity at discounts up to 91% off on-demand prices but can be preempted at any time with no SLA. Preemption follows an ordered sequence: first Compute Engine sets metadata preempted to TRUE, then sends an ACPI G2 Soft Off signal, then allows a best-effort shutdown period up to 30 seconds, then forces power off and applies STOP or DELETE. A custom machine type provisions cores independently of memory to avoid paying for unused resources and to limit per-core licensing costs.
Exam tip: Choose GKE for stateful services needing persistent volumes and Kubernetes control; choose Cloud Run for stateless, request-driven services needing automatic scaling and pay-per-use billing.
OS Login links Linux user accounts to Google identities so SSH access is controlled by IAM. OS Login is enabled by setting metadata key enable-oslogin to TRUE in project metadata for all VMs or instance metadata for one VM. Instance metadata takes precedence over project metadata, so enable-oslogin=FALSE on one VM disables it even when the project enables it. Once enabled, the VM fetches keys and permissions from the OS Login service on every login instead of using SSH keys in metadata.
IAM roles decide login privilege, where roles/compute.osLogin grants regular login without sudo and roles/compute.osAdminLogin grants administrator login with sudo. Project grants apply to all OS Login VMs, instance grants apply to one VM, and folder or organization grants flow down. Connections may also need IAP tunnel, service-account-user, and external-user roles, plus enable-oslogin-2fa and Google 2-Step Verification for two-factor authentication. VM access also uses service accounts for API calls, custom metadata for configuration, startup scripts that run on every start and reboot, and Shielded VM and Confidential Computing protections.
VM Manager provides centralized patching, inventory, and vulnerability reporting through the OS Config API and OS Config agent. The project needs the OS Config API enabled and enable-osconfig=TRUE in metadata, plus a service account and network path to osconfig.googleapis.com. A patch job is a one-time update run and a patch deployment adds a schedule that creates jobs automatically, using native tools such as yum, apt, and Windows Update Agent. Once the agent reports inventory, the service compares packages against vendor sources and shows findings in the console, Security Command Center, and Cloud Asset Inventory.
Deploying to GKE means storing images centrally, declaring desired state, and exposing it through a stable Service. A Deployment declares replicas, selector labels, Pod template labels, and container image, and the selector must match the template labels or Pods will not associate. A Service groups changing Pod endpoints behind a stable IP and port, where ClusterIP is internal only, NodePort opens the same port on every node, and LoadBalancer adds a cloud load balancer with an external IP. For HTTP routing by host or path, Ingress with HTTP(S) load balancing is recommended.
Connecting kubectl to GKE requires the Google Cloud CLI, the kubectl component, and cluster credentials. Initialization with gcloud init sets account, project, and location, while get-credentials fetches the endpoint and adds a context to $HOME/.kube/config. Traffic flows from kubectl to the kube-apiserver over TCP port 443, either through the Connect gateway where the agent forwards to the private endpoint, or directly to the private endpoint inside the VPC. GKE clusters run as Autopilot for minimal node planning or Standard for direct node control, as zonal with one control plane or regional with replicated control planes, and images are pulled securely from Artifact Registry with IAM and scopes.
Serverless event workloads run on Cloud Run and Cloud Run functions without servers and receive Google Cloud events through Eventarc. A Cloud Run service is an HTTPS endpoint on a *.run.app domain that scales to zero, with minimum-instances for warm capacity and maximum-instances for cost control. An Eventarc trigger listens for one event type and sends matches as HTTP POST in CloudEvents format, such as Pub/Sub google.pubsub.topic.publish or Storage google.cloud.storage.object.v1.finalized. Data flows from publisher to topic through the trigger to the function, and private services require the Cloud Run Invoker role or delivery fails.
Continue with this study guide: 2.1 Planning and implementing compute resources
2.2 Planning and implementing storage and data solutions is choosing databases, messaging, processing, caching, and object storage that fit structure, scale, latency, and location needs. Relational services enforce tables and SQL for transactions, non-relational services trade fixed schemas for automatic scale, and messaging plus processing moves data toward analytics. Caching then speeds repeated reads in front of slower databases.
Relational database services store data in tables with enforced schemas for online transaction processing (OLTP). Cloud SQL (opens in a new tab) provides managed MySQL, PostgreSQL, and SQL Server with automated backups, point-in-time recovery, high availability, and cross-region replicas. Spanner is the global-scale relational service that keeps transactional data strongly consistent across regions with GoogleSQL support. AlloyDB for PostgreSQL is the PostgreSQL-compatible service for demanding transactional workloads, while AlloyDB Omni runs outside Google Cloud.
Firestore is the scalable NoSQL document database built for automatic scaling and application development, and the successor to Datastore. A kind maps to a table, a document or entity maps to a row, a field or property maps to a column, and a document ID maps to a primary key. Entities of the same kind can have different properties, queries use prebuilt indexes, and transactions are atomic with strong consistency, but joins and inequality filtering on multiple properties are not supported. Bigtable fits data that is not highly structured and workloads that do not require ACID transactions, with replication across zones and backup by export to Cloud Storage.
| Service | Data model and role | Choose when |
|---|---|---|
| Cloud SQL | Managed MySQL, PostgreSQL, SQL Server tables | The workload needs full SQL for OLTP with backups and replicas |
| Spanner and AlloyDB for PostgreSQL | Global-scale or high-performance relational tables | The workload needs strong consistency at global scale or demanding PostgreSQL performance |
| Firestore | Schemaless NoSQL documents | The application needs automatic scaling with flexible fields |
| Bigtable | NoSQL for less-structured data | The workload does not need ACID transactions |
Low-latency caching with Memorystore for Memcached keeps frequently used data in memory instead of on disk. Data flows from the primary database into the cache on first read, and later reads flow from the cache to the application to reduce database load. A Memcached instance can contain one or many nodes with the same memory and vCPUs, keys are divided across nodes with no replication, and the application must run in the same region as the instance.
The analytics pipeline separates durable messaging, in-flight transformation, and SQL analysis into services that hand data forward in order. Producers publish events once to a Pub/Sub topic or to Google Cloud Managed Service for Apache Kafka, and consumers subscribe independently so ingestion keeps running when downstream jobs restart. A Dataflow job built with Apache Beam SDKs for Java, Python, or Go then reads from the topic, applies filtering and mapping, and writes processed rows to BigQuery. BigQuery is the central warehouse where curated data lands for large-scale SQL analysis.
In an extract, load, transform (ELT) flow, data is loaded into BigQuery first and then transformed with SQL, views, and routines. The BigQuery Data Transfer Service automates recurring scheduled loads from sources such as software-as-a-service apps, Cloud Storage, Amazon S3, Google Ads, and YouTube reports. Choose messaging when producers and consumers must operate independently, choose Dataflow when data must be cleaned or joined in motion, and choose BigQuery when the goal is centralized SQL analytics. Managed Service for Apache Spark runs Hadoop, Spark, and Flink jobs that read from Cloud Storage and write to BigQuery, favoring many ephemeral clusters over one monolithic cluster.
Shared file storage provides a network file system that multiple clients mount at the same time. Filestore is the managed service for general shared workloads reached over a Virtual Private Cloud (VPC) network through private services access. Google Cloud NetApp Volumes is the enterprise file service for demanding workloads such as high-performance computing and large migrations with intensive metadata operations. Filestore on Shared VPC requires first creating a peering address range, then connecting to servicenetworking.googleapis.com, and only after that creating the instance. NetApp Volumes organizes capacity into storage pools before volumes are created.
Cloud Storage classes price object data by access frequency. Standard storage fits frequently read data, Nearline storage also fits frequently accessed data, while Coldline storage and Archive storage fit infrequent and long-term retention data at lower storage cost. Autoclass moves each object automatically based on its own pattern: every new object starts in Standard, unaccessed objects cool toward Nearline by default or optionally Archive, and reading object data moves the object back to Standard. Object Lifecycle Management is the alternative fixed-rule policy that deletes objects or moves them to colder classes, such as removing noncurrent versions.
Location controls latency, egress cost, and residency. Data flows from a bucket in one region to compute in another region when separated, creating cross-region egress charges and added latency. The fix is bucket relocation so the bucket and services sit in the same region. Growth signals such as rising stored bytes often mean noncurrent versions are retained, while throttling signals such as spikes in 429 errors mean request rates exceed sustainable patterns.
Exam tip: In Autoclass, reading object data returns the object to Standard storage, but reading or editing object metadata does not trigger a transition.
Loading data means staging files in Cloud Storage and moving them into warehouses with the right tool for source and size. The gcloud storage cp command copies files and folders into buckets, with --recursive to walk subdirectories and wildcards to select groups. The gcloud storage rsync command synchronizes a directory, with --delete-unmatched-destination-objects to make an exact mirror. Large transfers are resumable based on bytes already received. Loads from Cloud Storage into BigQuery move Parquet, JSON, and CSV files into tables, with truncate to replace or append to add, while Storage Transfer Service moves large amounts over 1 TiB from Amazon S3, Azure Blob, on-premises, URLs, or HDFS into Cloud Storage with parallelism, retries, and TLS 1.3.
Multi-region redundancy keeps data available for global reads and regional failover. Cloud SQL high availability places primary and standby in two zones of one region with automatic failover inside that region. If the whole region fails, recovery requires a cross-region read replica that synchronizes asynchronously from the primary. Disaster recovery (opens in a new tab) follows an ordered sequence: first the primary region fails, then the team declares disaster, then the replica is promoted to the new primary, then clients are reconfigured. Promotion is manual and recovery point may be non-zero because the replica can lag.
Managed multi-regional services distribute data automatically without manual promotion, including Spanner, Cloud Storage, Datastore, and Firestore. BigQuery stores data synchronously in two zones of the specified region, and cross-region replication creates a continuous read-only copy near consumers but is not a disaster recovery plan for total outages. Dual-region and multi-region Cloud Storage buckets keep one active-active bucket redundant across places with a designed recovery time of zero. Default replication targets 99.9% of new objects within one hour and 100% within 12 hours, while Turbo replication for dual-region buckets targets 100% within a 15-minute recovery point objective.
Continue with this study guide: 2.2 Planning and implementing storage and data solutions
2.3 Planning and implementing networking resources is building private networks, firewall controls, hybrid connections, load balancing, and internet routing paths. A Virtual Private Cloud (VPC) (opens in a new tab) network carries internal traffic between resources, while each subnetwork holds addresses for one region. Firewall policies, peering, VPN, Interconnect, load balancers, and service tiers then decide who can reach each workload and how traffic travels.
Auto mode and custom mode VPC networks differ in how subnetworks are created. An auto mode network automatically creates one subnetwork in every region and gains a new one when Google opens a new region. A custom mode network starts with no subnetworks, so each subnetwork is created manually with 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 |
Regional subnet planning assigns a non-overlapping CIDR range to each region so routing does not conflict with the same VPC, peered networks, or on-premises networks. Planning also reserves space for future regions, test and production separation, and private service connections. Every subnet reserves 4 addresses, so usable addresses equal total addresses minus 4. A Shared VPC architecture places the central network in one host project and serves workloads in service projects, where the Shared VPC Admin enables the host, attaches service projects, and grants Network User access. The subnet belongs to the host project while the instance belongs to the service project, and the instance must be in the same region as the shared subnet.
Cloud Next Generation Firewall (Cloud NGFW) policies group firewall rules and attach them to the right networks. A firewall policy is the container and a firewall policy rule is the unit that allows, denies, or inspects traffic. Hierarchical firewall policies are created at organization or folder level and flow down to matching VPC networks, global network firewall policies apply to all regions of attached VPCs in the same project, and regional network firewall policies apply to one region. A policy has no effect until association is complete.
Every rule applies to either ingress or egress, never both. Ingress controls inbound packets arriving at the target and requires a source, while egress controls outbound packets sent from the target and requires a destination. Protocols and ports narrow the match, priority is a unique integer where lower means evaluated first, and stateful filtering remembers allowed connections so return packets are permitted automatically. Targets narrow rules to VM instances by default or managed Envoy proxies for internal load balancers.
Service accounts and secure tags let rules follow workloads as addresses change. A rule can name a source service account for the initiator and a destination service account for the receiver, so only matching identities are affected. Secure tags are IAM-governed tags bound to instances with purpose GCE_FIREWALL, while network tags are simple strings without access controls. Secure tags work in hierarchical, global, and regional policies, while network tags work only in VPC firewall rules.
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.
Hybrid 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. Traffic between peered networks stays internal, each network keeps its own subnets and firewall rules, and a VM reaches a peered VM by primary internal IP or alias ranges.
Peering is non-transitive, so if A peers with B and B peers with C, A does not automatically reach C. Each pair needing communication requires its own peering. A static route next hop must be a primary internal IP in the route own VPC, not an alias or peered-network address. Cloud VPN sends traffic through IPsec tunnels with Border Gateway Protocol (BGP) dynamic routing, where each autonomous system uses a unique number and peers can use MD5 authentication. High Availability VPN scales with Equal-Cost Multi-Path routing and carries a 99.99% availability SLA, but has internet egress billing, higher latency, limited per-tunnel throughput, and lower MTU from encapsulation.
Cloud Interconnect (opens in a new tab) creates a direct physical connection without traveling over the public internet. Traffic flows from the on-premises router across a private circuit to the Google edge and then into the VPC. Dedicated Interconnect uses a direct circuit in a colocation facility for very high volume, while Partner Interconnect uses a partner provider when direct access is unavailable or needs are variable. Both depend on Cloud Router to exchange routes with BGP after the VLAN attachment connects the circuit to the VPC.
Cloud Load Balancing places backends behind a single address that distributes client traffic. Global balancing serves multi-region workloads by creating backends near arriving traffic and failing over to another region if one region fails. The cross-region internal Application Load Balancer is a proxy-based Layer 7 Envoy balancer where traffic flows from the forwarding rule to the closest backend and VIPs from multiple regions can share one backend service. Regional balancing keeps all backends in one region for localized traffic, where the regional internal Application Load Balancer auto-allocates Envoy proxies and clients are limited by the global-access setting.
Layer 7 and Layer 4 handling match routing to the OSI layer. Application Load Balancer terminates HTTP/HTTPS and routes by host name and path through a URL map with host rules and path matchers. Layer 4 proxy and passthrough services handle TCP/UDP by IP, port, and session affinity without reading HTTP content. Backend services group instance groups with health checks, where traffic flows from URL map to backend service to backends in different zones. Health checks probe application responsiveness because a VM can be RUNNING while its app has crashed. The frontend chain runs from forwarding rule to target proxy to URL map to backend service, with HTTP on port 80 and HTTPS on port 443 requiring certificates.
Network Service Tiers controls which network carries traffic between the internet and Google Cloud. Premium Tier carries traffic on Google private fiber with over 200 points of presence, entering near the user and exiting near the destination for low, consistent latency. Standard Tier carries traffic over public internet transit, entering near the resource region and exiting in the hosting region for lower cost but higher, variable latency. Premium Tier is the default, is required for multi-region backends and global external IPs, and offers 99.99% VM uptime SLA, while Standard Tier offers 99.9% and 200 GB free per month per region.
| Tier | Routing path | Performance characteristic |
|---|---|---|
| Premium Tier | Traffic travels 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 |
Continue with this study guide: 2.3 Planning and implementing networking resources
2.4 Planning and implementing resources through infrastructure as code is defining Google Cloud infrastructure and applications in versioned files instead of manual clicks. Infrastructure as Code (IaC) (opens in a new tab) stores the desired end state in code, so the same files can create matching development, test, and production environments. Different tools cover platform resources, Kubernetes applications, and automated deployment workflows.
The Terraform lifecycle through Infrastructure Manager (Infra Manager) is the ordered path from configuration code to running resources. Infrastructure Manager is the managed service that runs Terraform configurations, also called blueprints, without a separate automation server. Configuration data flows from storage to an automated build job and then to live resources, logs flow to a managed bucket, and deployment metadata flows to storage for queries.
terraform init to install providers and terraform validate to check syntax.terraform apply to create resources, and terraform destroy when deleting.Infrastructure Manager stores Terraform state automatically as the source of truth per deployment and revision, including logs, configuration, resource lists, and state files. Because it handles state, backend blocks should not be defined, templating of Terraform is not supported, and provisioners are not recommended. A GitOps flow can run terraform plan on feature branches and apply only after merge. Fabric FAST blueprints are standardized Terraform packages for landing zones, where a module reuses configuration for one capability and a blueprint packages modules plus policy for a full foundation.
| Option | What it contains | Use it when |
|---|---|---|
| Module | Reusable configuration files for a logical group of resources | A single capability needs to be reused consistently across environments |
| Blueprint | Deployable modules plus policy and documentation for an opinionated solution | A full landing zone or foundation needs to be deployed the same way every time |
Exam tip: Do not define a backend block when deploying with Infrastructure Manager, because Infrastructure Manager stores the state file for each deployment and revision itself.
Config Connector is a Kubernetes extension that manages Google Cloud resources through declarative YAML with Custom Resource Definitions. Each object such as kind: SQLInstance creates a matching cloud resource, and the Kubernetes object remains the single source of truth. If the cloud resource drifts, Config Connector detects and corrects it. The spec field defines desired state written by the user, such as databaseVersion, while the status field reports observed state written by Config Connector, such as connectionName and ready condition. Data flows from applied YAML to Config Connector to Google Cloud, while observations flow back into status.
Object metadata maps Kubernetes organization to Google Cloud organization. Name defaults to the cloud resource name unless spec.resourceID specifies another, namespace determines the containing project, labels propagate to the cloud resource plus a managed-by-cnrm label, and annotations define extra actions. Access is controlled by Kubernetes RBAC, where permission to create the object grants ability to create the cloud resource. Helm fills the application layer by packaging Kubernetes manifests as versioned charts, where templates define common manifests with placeholders and values files supply per-environment settings for development, test, and production.
Safe updates follow immutable infrastructure, where a deployed VM, cluster, or firewall rule is never tweaked directly. The update is made to configuration in source control, tested through CI/CD, and fully redeployed, which avoids drift and snowflake servers. A deployment update either reapplies the original configuration to correct drift or applies a new configuration to implement change. Prerequisites include enabling Infrastructure Manager, granting caller roles/config.admin, providing a service account with resource permissions, identifying the deployment and configuration, avoiding sensitive data, confirming quota, and completing data migration. Preview shows proposed changes before apply, and transient init or apply failures are retried automatically.
| Tool | What it manages | Choose when |
|---|---|---|
| HashiCorp Terraform | Google Cloud infrastructure including VMs, networks, storage, and GKE clusters | The need is to provision platform resources with versioned configuration |
| Infrastructure Manager | Deployment of Terraform configurations on Google Cloud | The need is to deploy Terraform programmatically using a Google Cloud service |
| Helm | Kubernetes applications packaged as charts with templates and values files | The need is to package and deploy workloads inside a cluster across environments |
Choose Compute Engine when the workload needs full operating system control, custom kernels, or specialized GPUs and TPUs, choose Google Kubernetes Engine (GKE) when complex microservices need Kubernetes orchestration, custom networking, or persistent storage, and choose Cloud Run when stateless, request-driven services need minimal operations with automatic scaling and scale to zero. Google manages the GKE control plane while the user manages nodes, networking, scaling, and upgrades, and Cloud Run bills pay-per-use rounded up to the nearest 100 milliseconds.
Zonal Persistent Disk keeps one copy in one zone and the virtual machine must run in that same zone, so recovery from zone loss depends on snapshots. Regional Persistent Disk keeps two synchronously replicated copies in two zones of the same region, so a virtual machine in the surviving zone can continue without a restore. Google Cloud Hyperdisk (Hyperdisk) lets IOPS, throughput, and capacity scale independently, with Balanced for general use, Extreme for up to 350,000 IOPS, Throughput and ML for sequential throughput, and Balanced High Availability for synchronous replication across two zones.
An auto mode network automatically creates one subnetwork in every region and gains a new one when Google opens a new region, while a custom mode network starts with no subnetworks so each subnetwork is created manually with its region and IPv4 range. All subnetworks in the same Virtual Private Cloud (VPC) network can still reach each other because the Virtual Private Cloud (VPC) is one global network.
Premium Tier carries traffic on Google private fiber with over 200 points of presence, entering near the user and exiting near the destination for low, consistent latency, while Standard Tier carries traffic over public internet transit, entering near the resource region and exiting in the hosting region for lower cost but higher, variable latency. Premium Tier is the default and is required for multi-region backends and global external IPs.
Last updated