Prepare and test your skills
Prepare and test your skills
Integrating solutions with existing systems connects new Google Cloud applications to on-premises and other-cloud systems without disrupting operations. The design uses Apigee (opens in a new tab) as the central API gateway, Private Service Connect for private paths, and hybrid network links to carry traffic. Together these controls decide where traffic enters, how it is inspected, and how workloads move in phases.
Gated ingress through Apigee is a design where all inbound API calls must pass through a managed gateway before reaching a backend. Apigee provides controlled entry, authentication and policy enforcement, rate limiting to protect backends from overload, and analytics for visibility into API use. This central point lets teams enforce the same policy even when backends live in different environments.
Left-to-right process flow from remote network over VPN or Interconnect to a PSC endpoint linked to Apigee, then to an internal load balancer and application APIs, with direct bypass blocked.
Traffic in the gated path flows in a fixed order from the remote network to the application. Traffic flows from on-premises or another cloud over Cloud VPN or Cloud Interconnect to a Private Service Connect endpoint in the application Virtual Private Cloud (VPC) that is linked to the Apigee VPC. Traffic then flows from that endpoint to an internal load balancer in the application VPC, and the load balancer forwards the request to the application APIs. Firewall rules filter traffic in the application VPC, and hiding the backend subnet from the remote network prevents clients from bypassing the endpoint and gateway.
When inspection outside the application VPC is required, a separate transit VPC sits between hybrid connectivity and the application VPC. Traffic flows from the remote environment through the transit VPC to the Private Service Connect endpoint, where firewalls perform deep packet inspection, then continues to the internal load balancer and application. Multiple Private Service Connect endpoints can exist in the same VPC, with Cloud Router and firewall rules selecting paths and global access providing failover.
Apigee Adapter for Envoy is a small-footprint gateway that runs close to backend applications on premises or in another cloud. The adapter avoids deploying a full Apigee installation in every data center, which would be large because of features such as key management, monetization, and analytics. The adapter instead splits work between a management plane on Google Cloud and a data plane that runs remotely with the Envoy proxy and the Apigee Remote Service.
A client request moves through the adapter data plane in a fixed order:
Running near the backend keeps API traffic inside enterprise-approved boundaries, reduces latency for nearby services, and still provides Apigee analytics. The adapter still depends on the management plane for control functions including OAuth token generation, API key verification, quota enforcement, and sending analytics data. First-time API key checks and OAuth token generation require connectivity, while later key checks can use a cache.
Bidirectional integration with Private Service Connect lets applications in different environments call each other as private APIs without using the public internet or overlapping addresses. One direction uses an endpoint to expose a service, while the other uses an interface to reach back. The two paths are independent connections for different purposes, not two directions of one connection.
| Option | Direction of access | Purpose |
|---|---|---|
| Private Service Connect endpoint | Traffic flows from consumer VPC to a service published in the producer VPC | Expose Google APIs such as Cloud Storage or BigQuery, or an application backend, as a private address |
| Private Service Connect interface | Traffic flows from a producer virtual machine (VM) into the consumer VPC and beyond | Let a Google Cloud application reach consumer resources or on-premises sources reachable through that VPC |
A Private Service Connect interface is a network interface attached to the consumer or transit VPC but shared with a producer VM, while keeping producer and consumer roles separate. With that interface, the producer VM can access consumer resources as if they were local. When the consumer VPC belongs to an external organization, security must be defined in the guest operating system of the interface VM, or another approach must be chosen.
Hybrid connectivity links on-premises, Google Cloud, and other clouds, so bandwidth, latency, packet loss, and jitter directly affect performance. Cloud VPN uses IPsec over the public internet for lower cost and lower bandwidth needs, while Partner Interconnect provides private transport through a provider for larger transfers normally under 10 Gbps. Dedicated Interconnect provides a direct private connection for 10 Gbps or more where the team manages routers, and Cross-Cloud Interconnect provides managed private connectivity between clouds.
High-availability (HA) VPN securely connects an on-premises network to a VPC through IPsec tunnels. An HA VPN gateway receives two external IP addresses, one per interface, and each interface supports multiple tunnels. Based on topology, HA VPN can provide 99.99% or 99.9% availability, while a gateway with only one active interface provides no availability SLA. Cloud Router exchanges routes with the other side through Border Gateway Protocol for VPN and Interconnect attachments.
Protection combines network and perimeter controls. VPC Service Controls defines service perimeters at the project or VPC level to reduce data exfiltration, and perimeters can extend over authorized VPN or Interconnect. Hybrid DNS keeps Cloud DNS for Google Cloud names and on-premises servers for local names, with one VPC forwarding to on-premises and other VPCs using DNS peering to that VPC. Shared VPC lets multiple projects share one VPC, and Network Intelligence Center provides visibility for troubleshooting.
Exam tip: Cloud VPN encrypts traffic by default, while Cloud Interconnect requires HA VPN over Cloud Interconnect or MACsec for encryption.
Continue with this study guide: Integrating solutions with existing systems
Assessing and migrating systems and data moves existing servers, databases, and datasets to Google Cloud with correct sizing, cost estimates, and validation. Google Cloud Migration Center (opens in a new tab) is the central service that collects asset data, recommends target sizes, and generates a Total Cost of Ownership (TCO) report. Complete discovery prevents wrong assumptions about scope, performance, and cost before waves are planned.
Bulk data transfer options move large object and analytics datasets to Google Cloud without manual copy handling. Storage Transfer Service is the managed service for online object moves to Cloud Storage from Amazon S3, Azure Blob Storage, Data Lake, or on-premises files. Data flows from the source system to the destination Cloud Storage bucket, with the service managing scheduling, retry, and throughput. Choose this option when data is reachable over the network and the target is object storage.
Transfer Appliance is a physical appliance for very large offline moves that would saturate wide-area bandwidth. Data is copied from the source to the appliance on premises, the appliance is shipped to Google, and Google uploads the data to Cloud Storage. BigQuery Data Transfer Service automates recurring loads into BigQuery from software-as-a-service apps, Teradata, or Amazon Redshift, so analytics tables stay current without custom pipelines.
| Transfer option | Source to destination | Choose when |
|---|---|---|
| Storage Transfer Service | Online sources to Cloud Storage | Data is reachable online and target is object storage |
| Transfer Appliance | On-premises data via shipped hardware to Google Cloud | Data exceeds 20 TB up to 1 petabyte and network transfer would disrupt operations |
| BigQuery Data Transfer Service | SaaS apps, Teradata, or Redshift to BigQuery | Analytics data needs scheduled loading into BigQuery |
Exam tip: Choose Storage Transfer Service for online transfers and Transfer Appliance when offline shipment is needed for data over 20 TB up to 1 petabyte.
Database migration with built-in backups moves relational data to Cloud SQL or AlloyDB through an export file in Cloud Storage that is restored to the target instance. This dump-and-load path is used for one-time moves when significant downtime is acceptable and sources are relatively static. Managed services around this flow provide secure transfer, a common monitored interface, and lower cost than custom migration servers.
PostgreSQL moves use pg_dump and pg_dumpall with specific limits that affect planning. Use built-in utilities for databases of 500 GB or less, because larger dumps need substantial disk and memory. The pg_dump tool does not include users and roles, so use pg_dumpall for those accounts. Cloud Storage supports a maximum single-object size of 5 terabytes, so break larger exports into segments. When dumping, compress data to reduce cost and time, and after restore monitor pg_restore output, review logs, and run counts and queries to verify integrity.
Data Validation Tool is an open source Python tool that verifies correctness after migration and before cutover. The tool compares source and destination tables across different engines to check row counts, schemas, and values. Running validation before production cutover catches missing rows or transformation errors while fallback is still possible. Related tools include HarbourBridge for Spanner, HBase tools for Bigtable, and a Hive-to-BigQuery framework.
Discovery with Migration Center builds a complete inventory through automated collectors and imports from other clouds. For Amazon RDS or Aurora, the inventory includes engine version, CPUs, memory, topology, backup policies, parameters, and customizations. For batch platforms, the inventory records data infrastructure layers, data pipelines and their dependencies, and datasets those pipelines read. Automated discovery is preferred because manual spreadsheets cause errors in scope and sizing.
Resource evaluation turns inventory and performance data into a correctly sized target. Benchmarking measures current workload performance to estimate target needs, with special attention to disk size and throughput, IOPS, and vCPU count because Cloud SQL disk performance depends on disk size and vCPUs. To reduce latency, transfer cost, and I/O between layers, store data in the same zone where it is processed while keeping storage and compute on separate platforms. Migration Center then produces size recommendations and a TCO report for servers and databases.
Strategy selection chooses scheduled maintenance or continuous replication for each database. Scheduled maintenance requires an affordable cut-over window and offers lower cost with little refactoring, while continuous replication provides near-zero downtime for mission-critical databases with more planning. Waves then group dependent databases, apps, pipelines, and datasets so each wave can move and cut over in order, with small infrequent databases often using scheduled maintenance and critical databases using continuous replication.
Using migration methodologies, workload testing, network planning, and dependency planning turns an inventory into an ordered move to Google Cloud. The methodology decides how much each workload changes, testing proves it still works, networking connects source and target, and dependency mapping decides what moves together. Network design must come before moving workloads because it controls migration traffic, bandwidth, and hybrid-period costs.
Change-based migration strategies build on lift-and-shift and differ in what changes and how much time and skill they need. Mission-critical workloads use zero or near-zero downtime approaches to avoid data loss, while secondary workloads can accept a cutover window to stay simpler and cheaper. Migration should improve design, code quality, cost-efficiency, and use of built-in Google Cloud features rather than only lifting existing systems.
| Strategy | What changes | Tradeoff |
|---|---|---|
| Refactor | Modify workloads to use cloud scalability and high availability | Takes longer than rehost and needs new skills |
| Re-architect | Change how code functions, such as splitting a monolith into microservices | More complex than refactor and can introduce bugs needing retesting |
| Rebuild | Decommission the app and rewrite it as cloud-optimized | Takes longest and removes debt, but not for ready-made apps |
| Repurchase | Move to a SaaS equivalent such as Google Workspace | Easier on resources but can cost more with less control |
Migration Center (opens in a new tab) supports discovery and assessment, while managed migration services encapsulate move logic and monitoring. Tool choice depends on migration strategy, source and target environments, data and workload size, frequency of changes during migration, and availability of managed services. A large frequently changing dataset needs different tooling than a small static workload even with the same strategy.
Exam tip: Refactor restructures how workload code uses cloud features, while re-architect changes how that code functions, such as splitting a monolith into microservices.
A proof of concept (POC) is a small preliminary implementation that validates feasibility before full migration. The POC starts by defining scope and success criteria such as compatibility, minimal downtime, and performance demands. Findings are captured in migration plan documentation, and several POCs are iterated until criteria are met. Concrete examples include running the same logic against Cloud SQL, Spanner, Firestore, or Bigtable to compare benchmarks and costs, or using PerfKit Benchmarker to measure provisioning time, latency, and throughput.
A POC helps answer high-risk questions at low cost:
Validation of results is continuous before, during, and after migration. Before migration, existing test cases are refactored to reflect the Google Cloud target. During migration, results are validated at each milestone with integration tests between migrated and not-yet-migrated parts. After migration, baseline tests compare execution time and error rates, functional tests check inputs and outputs, and load tests gradually increase traffic to prove performance and scalability.
VPC planning establishes the global network container and regional subnets for migrated workloads. A VPC network is a global resource, while each subnet is bound to a single region and cannot be reused across regions. Because subnets cannot normally overlap, keeping the same addresses after a move typically requires a new VPC. Early choices for VPC Network Peering, Private Service Connect, private services access, and Shared VPC keep traffic private and let the landing zone support many apps.
Hybrid connectivity links the on-premises network to the VPC during migration. Cloud Interconnect extends the external network over a dedicated high-availability low-latency connection using VLAN attachments, which suits steady high-volume transfers. Cloud VPN extends the peer network over an IPsec tunnel using HA VPN tunnels and Cloud Routers, which suits faster provisioning and backup access. Establishing this connectivity first lets distributed components communicate, but VLAN attachments are regional, so leaving an attachment in a dismissed region forces extra region-to-region traffic.
Hybrid Subnets is a VPC routing mode that lets a VPC share a CIDR block with the connected on-premises network during migration. If a packet matches a hybrid route but its destination is not assigned in that VPC subnet, Google Cloud forwards it to the on-premises network. This lets migrated workloads in the VPC and remaining workloads on premises reach each other by internal IP without renumbering. After all workloads move, hybrid routing is disabled.
Enabling Hybrid Subnets requires three aligned pieces in the same region:
allow-cidr-routes-overlap flag, with the subnet range matching the on-premises CIDR.Dependency mapping records which application calls which database, service, network, or third-party system. The inventory is organized into a workload catalog scored on business criticality, dependency position, allowable downtime, and migration difficulty. Application dependencies include service calls, shared databases, and external middleware, while infrastructure dependencies include project structure, load balancers and DNS, identity controls, and secrets. Tightly coupled workloads break if a dependency is unavailable, while loosely coupled workloads tolerate outages through brokers, offline work, or retries.
A migration wave, also called a batch or sprint, is a group of workloads and data that move together because they share dependencies, teams, or operational needs. Grouping keeps a workload with the databases and queues it needs, so traffic does not cross between environments longer than necessary. Workloads are grouped by business value, deployment pattern, responsible teams, dependencies, refactoring effort, compliance and licensing, and allowable downtime. This grouping produces the technical design document, RACI matrix, and timeline such as a T-minus plan.
Sequencing moves foundational pieces first and builds skill with low-risk batches:
Good first-movers are non-mission-critical but commonly used, representative of wider patterns, owned by motivated teams, and able to tolerate a cutover window. Stateless microservices with no dependencies are preferred early because they avoid a phase where data lives partly in source and partly in Google Cloud. Workloads with strict latency must move with the databases they call frequently, and stateful databases must be synchronized for size, change rate, and consistency.
Determining software license implications and financial impact compares reusing existing licenses against paying for software as part of Google Cloud usage. Bring Your Own License (BYOL) reuses licenses already owned on Compute Engine (opens in a new tab), while license-included images bundle the charge into hourly usage. The choice affects long-term cost, scaling, and compliance work, and total cost must include network egress and cross-region traffic for data-intensive workloads.
Bring Your Own License is the practice of reusing software licenses already owned on Compute Engine instead of buying new on-demand licenses. Google Cloud adds no extra charge for bringing the image itself, but the customer still pays the vendor under existing agreements and pays Google Cloud for the infrastructure hosting the VMs. The customer must bring its own media, review licensing terms before importing, and contact the reseller for eligibility questions.
Sole-tenant nodes are physical servers dedicated to hosting VMs for only one project. VMs from the owning project run on one side of this hardware boundary, while other customers run on separate servers. Licenses with dedicated hardware requirements can only run on compliant dedicated hardware, so the imported image must run on sole-tenant nodes in a supported region. Node templates and groups can be configured to minimize the number of physical servers and cores that must be licensed.
| License scenario | Requires sole-tenant nodes | Portability path |
|---|---|---|
| Physical-core license with dedicated hardware requirement | Yes | Bring media and run imported image on sole-tenant nodes |
| Linux bring-your-own-subscription for RHEL or SLES | No | Bring subscription without sole-tenant nodes |
| Microsoft app with Software Assurance and License Mobility | No, when OS allows it | Run on default multi-tenant instances after verification |
The BYOL workflow moves licensed software from media to running VMs and then to vendor reporting in order. First prepare images per license agreements and activate licenses, then import virtual disks and create Compute Engine images. Then create a sole-tenant node template that defines configuration and a node group that reserves servers, provision VMs on the group from imported images, track physical core usage with reporting tools, and report consumption to the vendor.
Exam tip: Hardware-bound physical-core licenses require sole-tenant nodes, while RHEL or SLES subscription and Microsoft License Mobility for eligible apps do not.
License-included images and marketplace licenses collect the software charge together with infrastructure charges. With this model there is no separate vendor payment, instances can run on standard multi-tenant hosts, and cost grows and shrinks with running instances. Oracle Database@Google Cloud is a marketplace example where the license is purchased as part of the order and same-region transfer between apps and Exadata is included. This model fits variable workloads because autoscaling and rightsizing directly lower both infrastructure and license cost.
Microsoft portability depends on mobility rights or dedicated hardware. With Software Assurance with License Mobility, an eligible SQL Server license can run on default multi-tenant instances if the operating system does not itself require sole-tenant nodes. Verification is completed through Google or by naming Google LLC as Authorized Mobility Partner. Without mobility rights, SQL Server can only be brought to sole-tenant nodes when it meets Outsourcing Software Management Rights, with a minimum of four core licenses per physical processor and no upgrades to versions after October 1, 2019.
Per-core rules make Windows Server impact explicit. Windows Server per-core licensing is only available on sole-tenant nodes, where required licenses equal physical cores subject to a minimum of 8 per processor and 16 per server, sold in 2-packs. Standard edition grants two instances on the physical host, while Datacenter grants unlimited instances. Prerequisites also affect cost: new projects receive 72 CPUs of quota, so an n1-node-96-624 node type with 96 CPUs needs at least 192 CPUs for a minimum 2-node group, and host maintenance should use restart-in-place or migrate-within-node-group to minimize licensed servers.
Rightsizing matches provisioned capacity to measured utilization so resources are neither idle nor oversized. Compute Engine supports rightsizing during migration and after VMs run by applying machine type recommendations, while Google Cloud Observability provides monitoring data that reveals room to downsize. Practical actions include decommissioning idle resources, using custom machine types, deriving demand models from usage patterns, and running fault-tolerant work on Spot VMs at more than 80% lower cost than regular VMs. Because over-provisioning can also increase licensed capacity, rightsizing directly affects the TCO model.
Autoscaling adds and removes capacity with demand to avoid permanent over-provisioning. Compute Engine managed instance groups scale VM groups by policy with minimum and maximum counts, Google Kubernetes Engine (GKE) scales nodes, Pod replicas, and Pod requests through Cluster Autoscaler and Pod autoscalers, and Cloud Run scales serverless containers by incoming traffic. Choose managed instance groups for VM services, GKE autoscaling for containerized workloads needing node and Pod control, and Cloud Run when code should run without managing infrastructure.
Continuous auditing monitors spend, controls provisioning, and reduces waste to maintain compliance. Cloud Billing tracks current and projected spend through reports, exports to BigQuery, and budgets with alerts. Labels and tags support chargebacks and allocation analysis, while Cloud Asset Inventory compares current state to the documented state to find drift. Preventive controls include Organization Policy constraints, quotas on shared resources, least-privilege IAM, and centrally aggregated audit logs. SAP software runs only under BYOL, so customers must obtain SAP licenses from SAP or a reseller and use assessment data on cores, memory, and storage to plan both infrastructure and license scope.
Continue with this study guide: Determining software license implications and financial impact
A Private Service Connect endpoint exposes a service published in the producer Virtual Private Cloud (VPC) to a consumer VPC, while a Private Service Connect interface lets a producer virtual machine (VM) reach into a consumer VPC and beyond. Use an endpoint to expose Google APIs such as Cloud Storage or BigQuery or an application backend as a private address, and use an interface to let a Google Cloud application reach consumer resources or on-premises sources reachable through that VPC.
Storage Transfer Service is the managed service for online object moves to Cloud Storage when data is reachable over the network, while Transfer Appliance is a physical appliance for very large offline moves that would saturate wide-area bandwidth. Choose Storage Transfer Service when the target is object storage and sources include Amazon S3, Azure Blob Storage, Data Lake, or on-premises files, and choose Transfer Appliance when on-premises data exceeds 20 TB up to 1 petabyte and network transfer would disrupt operations.
Sole-tenant nodes, which are physical servers dedicated to hosting virtual machines for only one project, are required for physical-core licenses with dedicated hardware requirements. Linux bring-your-own-subscription for RHEL or SLES does not require sole-tenant nodes, and an eligible Microsoft application with Software Assurance and License Mobility can run on default multi-tenant instances when the operating system allows it.
Last updated