Prepare and test your skills

Prepare and test your skills

Shared responsibility and management scope describe how control of a workload is divided between Google Cloud and the customer. Infrastructure as a Service (IaaS) leaves the most control with the customer, while managed platforms shift operating system, runtime, and infrastructure work to Google. As abstraction increases, the customer manages less of the stack and Google manages more.
Three layered stacks compare responsibility: Compute Engine IaaS leaves OS and middleware to the customer, GKE and Cloud Run managed platforms leave only the workload to the customer, and Terraform Cloud SaaS is fully managed by Google.
Compute on Google Cloud means the virtualized processing power and memory used to run applications. These resources form a spectrum from raw virtual machine (VM) instances to fully managed platforms, covering VMs and bare metal, containers, batch processing, AI infrastructure, and high-performance computing (HPC) infrastructure. With Compute Engine (opens in a new tab), the customer can customize the operating system (OS), networking, and software stack. With managed platforms, Google handles most of the underlying infrastructure, so the customer focuses on the workload itself.
IaaS is the model with the most customer control and the most customer responsibility. The customer manages the OS, middleware, and applications, while Google manages virtualization, servers, storage, and networking. Google Cloud offers two IaaS paths: Compute Engine, which creates and runs VM and bare metal instances with choices of machine series, storage, accelerators, and guest OSs; and Google Cloud VMware Engine, which runs VMware workloads natively so teams can migrate without re-architecting. IaaS fits infrastructure architects, machine learning infrastructure engineers, and platform engineers who design custom environments from scratch.
Managed platforms balance control and ease of use by having Google manage infrastructure, OS, and runtime environments. The customer keeps responsibility for the workload that runs on top, which lowers day-to-day operational work but removes some low-level choices. Responsibilities divide by service:
| Service | What Google manages | What the customer manages |
|---|---|---|
| Batch | Provisions Compute Engine resources and manages their lifecycle | Scheduling, queuing, and running batch jobs at scale |
| Cloud Run | All serverless infrastructure management | Containers that respond to web requests or events |
| Cluster Director | Cluster infrastructure for AI, ML, or HPC | AI, ML, or HPC workloads |
| Cluster Toolkit | Deployment process through predefined blueprints | Resulting cluster infrastructure and software stack |
| Google Kubernetes Engine (GKE) | Kubernetes control plane | Container workloads |
For GKE, the customer also chooses a deployment mode: Autopilot gives more abstraction, while Standard keeps more infrastructure control.
Exam tip: In IaaS, the customer manages the OS, middleware, and applications; in managed platforms, Google also manages the OS and runtime, so the customer manages only the workload.
Workload governance uses the same core tools across IaaS and managed platforms. Identity and Access Management (IAM) controls who can create, modify, and delete compute instances, persistent disks, and networks. Cloud Monitoring collects metrics, events, and metadata to show deployment health, performance, and uptime through dashboards, charts, and alerts. Assured Workloads helps meet data residency, access control, support personnel, and regulatory requirements. Operational effort then grows with deployment scope (opens in a new tab): zonal deployments are simplest to operate, regional and multi-regional designs add monitoring, graceful handling of in-process requests, careful rollouts, and database synchronization, and hybrid and multicloud topologies add the most effort because tools and security controls differ across platforms.
Infrastructure as Code (IaC) provisions and manages infrastructure with code instead of manual console clicks or scripts. Teams write configuration files that declare the desired state, so the same files can create matching development, test, and production environments. This creates a single source of truth that can be reviewed, validated, audited, and reverted.
Staffing and technical skill requirements shrink as Google manages more layers of the stack. IaaS demands deep infrastructure skills because the customer owns the OS and middleware, while managed platforms let smaller or workload-focused teams operate without maintaining servers. Staffing plans must therefore match personnel capabilities to the management level of the chosen model.
IaaS staffing reflects maximum control and maximum responsibility. The customer team manages the OS, middleware, and applications, while Google manages virtualization, servers, storage, and networking. Compute Engine (opens in a new tab) requires staff who can select machine series, storage services, accelerators, and guest OSs and then patch, secure, and operate them. Google Cloud VMware Engine still requires staff who can manage VMs, guest OSs, and applications, even though Google manages the hardware and VMware platform. These environments are typically staffed by infrastructure architects, machine learning infrastructure engineers, and platform engineers who build environments from scratch.
Managed platforms shift operational work to Google so teams focus on workloads instead of servers. Google manages infrastructure, OS, and runtime, which reduces the depth of infrastructure expertise the customer must supply. The tradeoff is less control over low-level configuration in exchange for faster provisioning and lower ongoing maintenance.
| Model | Who manages OS and runtime | Example services | Operational effort |
|---|---|---|---|
| IaaS | Customer manages OS, middleware, and applications | Compute Engine, Google Cloud VMware Engine | Higher, team designs and maintains environment |
| Managed platforms | Google manages infrastructure, OS, and runtime | Batch, Cloud Run, Cluster Director, Cluster Toolkit, GKE | Lower, platform absorbs infrastructure work |
Batch removes direct VM management by provisioning resources and managing their lifecycle. Cloud Run removes server and cluster management for containerized web and event workloads. Cluster Director automates AI, ML, and HPC cluster setup with minimal overhead, while Cluster Toolkit abstracts deployment but leaves the resulting cluster to the customer. GKE splits the difference by having Google manage the control plane while the customer manages container workloads.
Exam tip: IaaS teams manage the OS, middleware, and applications; managed-platform teams focus on workloads because Google manages infrastructure, OS, and runtime.
Operational complexity and staffing needs grow with the number of locations and stacks a team must operate. A zonal architecture is easier to staff than an application running redundantly across zones or regions. Multi-location operation requires monitoring stacks in every location, handling in-process requests gracefully when a component fails, rolling out changes carefully, and synchronizing databases across locations. A global design with resources such as a global load balancer and global database can be simpler than many regional copies, but changes to global resources must be managed carefully. Hybrid and multicloud topologies (opens in a new tab) need the most staff because resources must be provisioned consistently across platforms, security tools are not standard, and teams can use tools like Terraform to reduce provisioning effort.
Compute skills alone do not staff a cloud team because every deployment also needs networking, storage, identity, monitoring, and compliance skills. The scope of each area depends on how much the chosen service model leaves to the customer.
Teams must also plan for quotas, since high-CPU or high-memory machines may require extra quota, and for personnel and overhead costs to design, build, and maintain the deployment.
Total cost of ownership (TCO) and flexibility pull in opposite directions across service models. IaaS offers the most customization but requires more spending on people and operations, while managed platforms lower long-term overhead by limiting low-level choices. SaaS goes furthest toward low overhead by delivering a finished application, such as Terraform Cloud for remote Terraform runs.
TCO is the full cost of running a workload, not just the hourly price of cloud resources. TCO includes provisioned resource costs plus personnel expenses and overhead to design, build, and maintain the deployment. The Google Cloud Pricing Calculator estimates resource costs, while the Well-Architected Framework guides careful utilization management. Fair comparison uses cost per unit of work, such as users served or requests processed, so a low resource price does not hide high operational effort. Other TCO drivers include:
Exam tip: Compare options by cost per unit of business value, such as requests processed, not by resource price alone, because operational effort is part of TCO.
IaaS provides the most flexibility because the customer controls the OS, middleware, and applications while Google manages virtualization, servers, storage, and networking. Compute Engine supports that flexibility with choices of machine series, storage services, accelerators such as graphics processing units (GPUs) or tensor processing units (TPUs), and guest OSs. Google Cloud VMware Engine offers a second IaaS path where Google manages hardware and the VMware platform while the customer keeps full control of VMs and applications. This freedom requires long-term budgeting for patching, securing, scaling, and staffing specialized infrastructure roles.
Managed platforms trade some customization for lower overhead and more predictable budgeting. Google manages infrastructure, OS, and runtime so teams focus on workloads instead of servers. Batch provisions and manages Compute Engine resources for large-scale jobs without direct VM management. Cloud Run runs containerized applications on serverless infrastructure in response to requests or events without managing servers or clusters. GKE manages the Kubernetes control plane while customers deploy workloads, with Autopilot mode abstracting more infrastructure and Standard mode preserving more control.
Deployment scope links availability choices directly to resource cost and operational effort. Zonal designs cost less and are simpler, while spreading across regions adds storage, transfer, and synchronization work. Global resources can simplify operations versus many regional copies, and hybrid or multicloud adds the most cost because provisioning and security differ by platform.
A comparison matrix of deployment archetypes arranged in rows, showing each archetype linked to its Google Cloud resource cost level and its operational complexity, plus staffing implications: zonal is cheapest and simplest, regional and multi-regional cost and complexity rise, global is medium cost, and hybrid/multicloud demands the most effort.
| Deployment archetype | Cost of Google Cloud resources | Operational complexity |
|---|---|---|
| Zonal | Low | Simpler than other archetypes |
| Regional | Medium | More complex than zonal |
| Multi-regional | High | More complex than regional |
| Global | Medium | Potentially simpler than multi-regional |
Choose zonal when low cost and simple operation matter more than surviving a zone failure. Choose regional, multi-regional, or global designs (opens in a new tab) when higher availability justifies redundant storage, cross-region transfer, and synchronized databases. A global load balancer requires Premium Tier networking, while a regional load balancer can use Standard Tier networking, so network tier is part of the cost decision.
The choice among models is a choice about who does the work. IaaS keeps control with the customer at the price of higher staffing, deeper skills, and larger TCO beyond resource bills. Managed platforms move OS, runtime, and lifecycle work to Google, lowering operational burden while narrowing customization and shifting staff toward workload, data, and delivery skills. Deployment scope then multiplies whichever model is chosen, so availability, location, and hybrid needs must be weighed alongside service-model flexibility.
In Infrastructure as a Service (IaaS), the customer manages the operating system, middleware, and applications while Google manages virtualization, servers, storage, and networking. In managed platforms, Google manages infrastructure, operating system, and runtime so the customer manages only the workload.
Total cost of ownership includes resource costs plus personnel expenses and overhead to design, build, and maintain the deployment, so fair comparison uses cost per unit of work such as users served or requests processed. Infrastructure as a Service (IaaS) offers the most customization with choices of machine series, storage services, accelerators, and guest operating systems but requires long-term budgeting for patching, securing, scaling, and staffing specialized roles, while managed platforms trade some customization for lower overhead and more predictable budgeting.
Infrastructure as a Service (IaaS) demands deep infrastructure skills because teams must select machine series, storage, accelerators, and guest operating systems and then patch, secure, and operate them, typically requiring infrastructure architects, machine learning infrastructure engineers, and platform engineers. Managed platforms reduce the depth of infrastructure expertise required because Google manages infrastructure, operating system, and runtime, allowing smaller or workload-focused teams to focus on workloads instead of servers.
A retail company is migrating its relational databases to Google Cloud and is comparing two deployment options:
During a budget review, the finance director notes that raw monthly infrastructure costs for Compute Engine appear lower than Cloud SQL. However, the cloud architect advises evaluating both long-term total cost of ownership (TCO) and customization requirements.
Which statement accurately compares the TCO and flexibility tradeoffs between these two cloud service models?
Long-term cloud budgets only need to account for direct monthly resource billing rates because neither IaaS nor PaaS requires budgeting for software licensing, personnel, or anticipated growth
Self-hosting on IaaS provides greater control over the operating system and custom machine configurations, whereas managed PaaS lowers long-term TCO by reducing operational overhead for patching, scaling, and disaster recovery
Self-hosting on IaaS eliminates ongoing personnel expenses through built-in database maintenance, whereas managed PaaS increases long-term TCO by requiring dedicated staff to manage physical hardware
Managed PaaS provides unrestricted root access to customize the underlying operating system and CPU instruction sets, while IaaS restricts deployments to fixed, unmodifiable server templates
Last updated