Prepare and test your skills

Prepare and test your skills

General cloud concepts explain why organizations adopt the cloud and how responsibilities and placement choices shape cost and control. The economic shift moves spending from owned hardware to usage-based billing, while service models and deployment archetypes define who manages what and where workloads run. Google Cloud (opens in a new tab) supports this shift with elastic capacity, shared pools of resources, and options from zonal to global availability.
The shift from capital expenditure (CapEx) to operational expenditure (OpEx) changes IT spending from large upfront purchases to ongoing pay-for-use charges. With CapEx, an organization buys servers, builds data centers, and pays for power, cooling, maintenance, and support before any workload runs. With OpEx, the organization provisions cloud capacity quickly and pays based on actual usage, so teams can start small, handle seasonal peaks, and avoid paying for idle servers.
Total cost of ownership (TCO) is the complete lifetime cost of a workload, including hardware, software, power, cooling, patching, scaling, security hardening, and disaster recovery. A TCO-aware view compares self-hosted data centers with managed cloud services that reduce operational work. When higher cloud spend is driven by revenue growth or transaction volume, further investment is justified, while wasteful use should be right-sized or re-architected.
Rapid elasticity is the ability to add or remove capacity automatically as demand changes, so performance stays stable without overprovisioning. Traffic flows from users to the autoscaling control and then to the added capacity, and the reverse happens when demand falls. Compute Engine managed instance groups (MIGs) scale groups of virtual machines (VMs) on CPU thresholds, Google Kubernetes Engine (GKE) scales nodes with Cluster Autoscaler and Pods with Horizontal and Vertical Pod Autoscalers, and Cloud Run scales serverless containers directly with incoming traffic.
On-demand self-service and shared resource pooling let teams provision and monitor their own resources without waiting for manual setup. The provider maintains large pools of compute, storage, and networking across zones and regions, and each customer draws from the pool on demand. Cost optimization lowers the OpEx rate while keeping flexibility through peer pricing choices:
Exam tip: Choose Spot VMs when the workload is fault-tolerant and flexible to save more than 80% compared with regular VMs; choose CUDs when usage is steady and predictable over one or three years.
Deployment archetypes (opens in a new tab) describe where public cloud workloads run and how placement affects availability, cost, and complexity. A public cloud runs entirely in one provider, a private cloud runs in a dedicated environment, hybrid combines on-premises with Google Cloud, and multicloud combines Google Cloud with other providers. Hybrid and multicloud add networking, monitoring, and security work because tools and controls are not standard across environments.
Google Cloud availability archetypes trade simplicity and cost for resilience:
| Archetype | Availability target | Outage resilience |
|---|---|---|
| Zonal | 99.9% | Low cost and simple, but recovery can take hours or days |
| Regional | 99.99% | Survives zone outages with near-zero recovery time objective (RTO) when replication is synchronous |
| Multi-regional and global | 99.999% | Survive zone and region outages with near-zero RTO when replication is synchronous |
The shared responsibility model divides security and maintenance between provider and customer. The provider secures the cloud infrastructure and supplies built-in high availability, backups, and scalability for managed services backed by service level agreements (SLAs). The customer secures data, configurations, access, and cross-platform controls placed in the cloud.
Cloud service models define how much of the technology stack the provider manages and how much control the customer keeps. As movement goes from Infrastructure as a Service (IaaS) to Platform as a Service (PaaS) to Software as a Service (SaaS), the provider manages more and the customer does less operational work. More customer control always means more management effort, while more provider management means faster use with less flexibility.
| Model | Provider manages | Customer manages | Choose when |
|---|---|---|---|
| Infrastructure as a Service (IaaS) | Physical hardware, virtualization, networking | Operating system, runtimes, data, applications | Custom workloads or migrated VMs need full control, such as Compute Engine |
| Platform as a Service (PaaS) | Hardware plus operating system, runtime, and scaling | Application code and data | Developers need to build apps fast without managing servers, such as App Engine or Cloud Run |
| Software as a Service (SaaS) | Almost everything, including the finished application | User settings and data inside the app | A ready-to-use app is needed, such as Gmail or Google Workspace |
Continue with this study guide: Explain general cloud concepts.
Transitioning to cloud infrastructure changes how applications grow, survive failures, release features, and consume budget. Scalability adjusts capacity, elasticity automates that adjustment, and reliability keeps services working through failures. Aligning cloud spending with business value (opens in a new tab) connects these operational gains to TCO, agility, and measurable business outcomes.
Scalability is the ability of a system to handle more or less work by adding or removing resources. A scalable web application serves one user or many users by adding capacity during peaks and removing unused capacity to save cost. Google Cloud supports scalability with Compute Engine VMs, GKE clusters, and data services such as BigQuery, Spanner, and Bigtable that hold performance steady at large scale, while Cloud Monitoring supplies the metrics for scaling decisions.
Elasticity is automated scalability that changes capacity without manual action when workloads rise or fall. Manual scaling requires people to watch utilization and adjust allocation, where overprovisioning wastes money and underprovisioning causes slowdowns. Compute Engine managed instance groups (MIGs) automate this change by managing VMs as one entity within minimum and maximum limits, adding VMs when CPU passes a threshold and removing them when it drops.
Reliability is the ability of an application to keep working despite component failures, with availability measuring the share of time a service is usable. Resilient designs use redundant compute, load-balanced traffic, and replicated data so one failure does not stop the service. Three design themes support resilience: automation of provisioning and deployment, loose coupling of independent components, and data-driven decisions from metrics and logs.
Exam tip: Scalability is the ability to add or remove resources, elasticity is automation of that change in response to demand, and reliability is continuing to function despite failures.
Total Cost of Ownership (TCO) is the full cost of running a workload over its lifetime, not just the upfront price. Moving to the cloud shifts spending from CapEx for owned servers and facilities to OpEx for consumed virtual machines, storage, and services. On-premises ownership also carries hidden costs for rack space, power, cooling, hardware amortization, and administrator salaries that shrink when the provider manages the physical data center.
Pay-as-you-go pricing bills only for time and capacity used, and autoscaling strengthens savings by removing idle capacity. Cloud Run scales serverless instances up with traffic and down to reduce cost, while GKE autoscaling tunes nodes and Pods together. Discounts further lower the rate for steady or flexible use, including CUDs for one-year or three-year commitments, sustained use discounts for consistent Compute Engine use, and Spot VMs at more than 80% lower cost for fault-tolerant work.
Business agility is the speed at which an organization experiments, releases features, and responds to market change. Self-service provisioning lets teams create test environments in hours without hardware orders, such as a three-to-six-month research project with on-demand high compute. Delivery practices such as DORA metrics for change failure rate, time to detect, and time to restore, plus Site Reliability Engineering (SRE) for automation and observability, shorten cycle times and reduce downtime.
Technical flexibility is the freedom to choose the right compute, storage, analytics, and application services for each need. Drivers for hybrid and multicloud adoption (opens in a new tab) include advanced analytics and AI, faster automated rollouts, high-level APIs, rapid provisioning, serverless elasticity, and global infrastructure. A multicloud approach extends choice and avoids vendor lock-in by using unique capabilities from each provider, but it requires weighing interoperability, consistent security, management complexity, and cross-cloud cost.
The CapEx-to-OpEx transition replaces fixed owned capacity with variable usage-based costs and changes how TCO is measured. CapEx pays upfront for physical infrastructure owned in advance, while OpEx pays as cloud resources are consumed. Quick TCO Estimator (opens in a new tab) in Migration Center compares these two cost structures side by side over one year and five years.
CapEx is spending on physical infrastructure that the organization owns, plans in advance, and buys for peak capacity. In an on-premises data center this includes servers, storage, routers, switches, cabling, and software licenses, plus ongoing rack space, power, cooling, maintenance, and IT staff salaries. Because hardware is purchased before use, much of this spending is fixed whether or not capacity is fully used.
OpEx is spending that follows actual consumption, so Google Cloud costs rise and fall with demand. Cloud TCO groups usage into Compute for virtual CPUs and memory time, Storage for disk types and capacity, Network for data movement, Operation for staff and management time, and operating system (OS) licenses. Autoscaling makes costs more variable by adding resources when traffic rises and removing them when it falls, which avoids paying for idle overprovisioned capacity.
| Dimension | CapEx on-premises | OpEx cloud |
|---|---|---|
| Payment timing | Upfront purchase before use, spread over hardware life | Ongoing charges based on resource usage |
| Cost behavior | Fixed capacity set by host count and admin ratios | Variable capacity adjusted by autoscaling and right-sizing |
| TCO inputs | Hardware, raw storage, network hardware, facilities, salaries | Compute, Storage, Network, Operation, OS licenses, migration, support, and discounts |
Exam tip: CapEx is tied to owning physical infrastructure and is spread over its hardware life; OpEx is tied to consuming cloud resources and is billed as usage occurs.
Financial agility is the ability to spend in small flexible steps and redirect money toward business value. Replacing large hardware purchases with OpEx removes the need to overbuy servers as a safety net, because extra capacity can be added and removed on demand. Leaders can then move capital from running data centers toward innovation and growth.
Elastic scaling carries capacity from the shared cloud pool to the workload when demand rises and returns it when demand falls. Compute Engine MIGs add and remove VMs within limits, GKE autoscaling adjusts nodes and Pods through Cluster Autoscaler and Pod autoscalers, and Cloud Run adjusts instances with incoming traffic. Self-service tools and reliable delivery practices turn these savings into innovation, because teams provision on their own while DORA metrics and SRE reduce failures without extra standby capacity.
On-premises TCO includes direct purchase costs plus indirect facility and staffing costs spread over hardware life. Modeling inputs include hardware life in years, raw storage price per GiB, monthly data center cost per host, IT operations salary, hosts per administrator, CPU and memory burst, vCPU-to-core ratio, and storage redundancy policy. A rigid data center often forces overprovisioning, which adds cost without value.
Google Cloud TCO is built from usage-based service costs plus migration, support, and discount effects. Network cost is commonly overlooked because on-premises routers and cabling are one-time purchases for full capacity, while cloud network utilization and egress are billed by use. Long-term control follows a fixed order: first monitor costs with Cloud Billing reports, labels, and budget alerts; then control provisioning with policies and ownership; then reduce costs with discounts, right-sizing, autoscaling, managed services, and data lifecycle management.
Private, hybrid, and multicloud choices match each workload to the right placement based on pattern, modernization goal, and integration need. A private environment keeps a workload in its existing data center, hybrid connects that environment to Google Cloud, and multicloud spreads workloads across providers. Planning a hybrid and multicloud strategy (opens in a new tab) helps weigh business effect, compliance, and complexity before placement.
Cloud-first hybrid integration launches new workloads in the public cloud while existing workloads stay where they are. A workload stays in classic private computing only when public cloud deployment is impossible for technical or organizational reasons. Cloud-first works best for selected workloads that gain the most, because migrating a costly existing workload can sometimes save more than placing a small new one in the cloud.
Mix-and-match decisions let different components of one stack follow different paths at the same time. An on-premises MySQL backend can be replatformed to managed Cloud SQL, frontend VMs can be refactored to containers on GKE Autopilot, and hardware load balancing can be replaced with Cloud Load Balancing and Google Cloud Armor. Traffic flows from users through Cloud Load Balancing and Google Cloud Armor to the container frontend, which then reads and writes data in Cloud SQL.
| Approach | What changes | Choose when |
|---|---|---|
| Rehost (lift and shift) | Move with little change | Few dependencies, refactoring is not worthwhile, or third-party software limits changes |
| Refactor (move and improve) | Untangle and improve for cloud | Dependencies, platform limits, inefficient resource use, or lack of automated deployment block efficient use |
| Rebuild (remove and replace) | Replace the application | Requirements changed, technology reached end of life, or license fees are no longer economical |
Regulatory and data residency rules about where data must be stored vary by industry and country. The most common sovereignty case is expansion into a country where the current provider has no local region, so the compliant solution is to use another provider that has one. Data classification and services such as Sensitive Data Protection help identify storage locations and movement limits for sensitive classes such as personally identifiable information (PII).
Security and workload sensitivity determine how much control and consistency is needed across environments. Each provider has its own security models, best practices, and shared responsibility boundaries, and separate monitoring tools create siloed visibility. Recommended practices include a unified security strategy, standardized design, layered defense-in-depth controls, continuous posture improvement, and centralized visibility with tools such as Security Command Center and cloud security posture management.
Business drivers must be weighed against tradeoffs in management, integration, and cost. Lasting drivers include sovereignty compliance, unique capabilities from each provider, avoiding vendor lock-in, and resilience by routing traffic to a second provider in the same region if one fails. Temporary drivers include short-term high-scale projects, multi-year transformation waves, and post-merger integration, where hybrid naturally appears while workloads move in waves.
Exam tip: Data sovereignty rules can force a multicloud choice: if the current cloud provider has no local region in a country, using another provider with a local region is the common compliant solution.
Private infrastructure is an organization's own computing environment, often an on-premises data center with direct control over systems and data. This control can help with security and compliance, but private environments may struggle to build advanced analytics and AI and may lack rapid provisioning. Private infrastructure often becomes one side of a hybrid architecture during migration.
Hybrid infrastructure combines private and public cloud so workloads run in both places and connect across a boundary. Multicloud infrastructure uses two or more public providers to gain sovereignty options, unique services, and resilience. Both add complexity in interoperability, cost visibility, and security, including larger attack surfaces and inconsistent models that require open practices and managed open-source services such as Kubernetes to contain.
| Model | Where workloads run | Best fit | Operational implication |
|---|---|---|---|
| Private | Own environment, often on-premises | Control, security, compliance, existing data | Limited scale and advanced capabilities; often joins hybrid |
| Hybrid | Private plus public cloud | Migration, temporary capacity, transformation, merger | Cross-environment networking, larger attack surface, complex data movement |
| Multicloud | Two or more public providers | Sovereignty, unique capabilities, lock-in avoidance, resilience | Management complexity, consistent-security work, skills and cost-visibility needs |
Network infrastructure terminology describes how names become routable locations and how physical transport carries traffic worldwide. An IP address is the numeric destination for packets, while Domain Name System (DNS) (opens in a new tab) translates human-friendly names into those addresses. In Google Cloud this translation works with load balancers and points of presence over fiber and subsea cables.
The global backbone is Google Cloud's worldwide network that carries traffic between regions and enables global load balancing. The backbone is built from owned fiber, public fiber, and subsea cables that move high volumes of data between continents with high throughput. This foundation gives global reach and lets traffic stay on Google's network until it is close to its destination.
Points of presence (PoPs) are network edge locations where customer traffic enters or leaves the Google network. An internet service provider (ISP) gives users initial internet access and carries traffic toward the nearest PoP. Traffic then flows from the user device to the PoP and across the private backbone to the destination region, which reduces public internet hops, lowers delay, and limits interception opportunities.
IP addresses are the routable identifiers that let traffic find a specific VM, load balancer, or service endpoint. Google Cloud uses internal private addresses inside a network and external addresses for traffic reachable from outside. Internal addresses come from subnets that segment address space by region, while static external addresses are regional resources used within the same region, including by regional forwarding rules.
DNS resolution returns the matching IP address when a user or application requests a domain name, so traffic flows from the requester to the correct resource. Internal DNS creates names for Compute Engine VMs and internal load balancers automatically. Cloud DNS serves public zones visible to the internet and private zones visible only inside a network, and it can forward private requests between on-premises and cloud networks in hybrid designs.
Exam tip: A Cloud DNS public zone is visible to the internet, while a Cloud DNS private zone is visible only within your network.
A region is a specific geographic location for deploying resources, and a zone is an isolated deployment area inside a region. Each region holds multiple zones so failure in one zone does not stop the others. Network edge data centers in more than 200 metropolitan areas bring traffic onto Google's network early and serve cached content close to users through Cloud CDN.
Latency is the round-trip delay a user experiences, and region choice affects only latency to the Compute Engine region, not last-mile latency. A rough rule is 1 millisecond per 100 kilometers of travel, with real paths 1.5 to 2 times map distance. Bandwidth affects how much data can move and what it costs, with three directions shaping high-availability design:
Google Cloud supports digital transformation with a worldwide foundation of data centers, fiber, undersea cables, and points of presence linked by a fast private network. This foundation lets applications scale like consumer services, place workloads near users, and move from older facilities to efficient cloud data centers. Geography and regions (opens in a new tab) organize this footprint from global network to region to zone.
Global infrastructure enables business agility by giving applications fast, redundant, and scalable transport for change. Customer traffic stays within Google's network until close to its destination, which reduces public hops, lowers latency, and limits attack opportunities. Redundancy of everything means no service depends on a single server, data center, or connection, while managed services such as Cloud Run, Compute Engine, and Firestore scale automatically with traffic.
Migrating on-premises workloads to Google Cloud can reduce energy use and associated emissions by 1.4 to 2 times compared with typical on-premises deployments. Cloud data centers are designed for high power usage effectiveness with renewable energy, climate-conscious cooling, and hardware optimization. Customers improve sustainability in the cloud by right-sizing VMs, using serverless services that scale to zero, and managing data lifecycles, which saves both energy and cost.
Regions are geographic areas that contain multiple zones, and zones are single deployment areas for resources such as Compute Engine VMs. Zonal resources live in one zone, regional resources spread across zones in one region, and multi-regional resources store data redundantly in at least two geographic locations. This hierarchy lets teams choose spread from one zone for testing to multiple zones and regions for production resilience and compliance.
Storage shows how zone protection works in practice. Regional Persistent Disk volumes replicate synchronously across two zones, Cloud Storage regional buckets replicate synchronously across zones in the region, and Filestore regional instances replicate across three zones. For larger failures, dual-region or multi-region buckets, multi-region Spanner, and snapshots to multi-regional storage survive extended loss of an entire region. Recovery planning ties this design to cost through recovery time objective (RTO) for maximum offline time and recovery point objective (RPO) for maximum data-loss window, where smaller targets generally cost more.
Exam tip: Use multiple zones in one region to survive a zone outage, and use multiple regions or multi-regional storage to survive the extended loss of an entire region.
The private network and points of presence move user traffic from the public internet onto Google's backbone early in the journey. Regions connect through owned fiber, public fiber, and undersea cables, while PoPs serve as global peering locations linked to most ISPs. Traffic flows from the user device to a nearby PoP and then within Google's network to the data center, so fewer public hops means lower latency and less exposure.
Low latency and high-speed delivery come from software-defined networking (SDN) and edge caching on top of that fiber foundation. SDN virtualizes the physical layer so software controls routing and load balancing, while edge caching places frequently used content near users. Security and availability are built in through defense in depth, where traffic passes through Google Front End (GFE) servers that help detect malicious requests and distributed denial-of-service (DDoS) attacks, plus firewalls, access control lists, automatic mirroring across locations, and restricted device access.
Capital expenditure (CapEx) is spending on physical infrastructure that is owned, planned in advance, and bought for peak capacity before any workload runs. Operational expenditure (OpEx) is spending on cloud capacity that is provisioned quickly and billed based on actual usage, so costs rise and fall with demand. Total cost of ownership (TCO) is the complete lifetime cost of a workload, including hardware, software, power, cooling, patching, scaling, security hardening, and disaster recovery.
Spot VMs give access to unused virtual machine capacity at a lower price for fault-tolerant, flexible workloads and can save more than 80 percent compared with regular virtual machines. Committed use discounts (CUDs) give discounts for committing to minimum usage or spend for one or three years when usage is steady and predictable.
Scalability is the ability of a system to handle more or less work by adding or removing resources. Elasticity is automated scalability that changes capacity without manual action when workloads rise or fall. Reliability is the ability of an application to keep working despite component failures, with availability measuring the share of time a service is usable.
As movement goes from Infrastructure as a Service (IaaS) to Platform as a Service (PaaS) to Software as a Service (SaaS), the provider manages more and the customer does less operational work. Infrastructure as a Service (IaaS) leaves the operating system, runtimes, data, and applications to the customer for full control, Platform as a Service (PaaS) leaves only application code and data to the customer for building apps fast without managing servers, and Software as a Service (SaaS) provides a finished application where the customer manages only user settings and data inside the app.
Last updated