Prepare and test your skills
Prepare and test your skills
Risk assessment and data-driven evaluation is the practice of finding technical and business risks before choosing an architecture and then using measured data to justify the choice. In Google Cloud, Migration Center (opens in a new tab) and assessment tools surface risks and dependencies, while Cloud Monitoring (opens in a new tab), BigQuery, and Cloud Billing (opens in a new tab) turn operations and spend data into key performance indicators (KPIs), dashboards, and cost-benefit analysis. Together these tools connect risk, performance, and cost to business value and long-term total cost of ownership.
Migration risk assessment evaluates what could go wrong during a move and whether a workload fits hybrid or multicloud (opens in a new tab) operation. The assessment covers application compatibility, pricing models, security features, interoperability, data privacy, compliance, consistency needs, and solutions across environments. Risks found in this assessment directly affect which workloads are chosen to migrate or run in the cloud. The factors used to evaluate migration risk include:
Technical difficulty is assessed separately from business risk and focuses on how hard the workload will be to move and run. The factors used to evaluate technical difficulty include:
To reduce risk, start with a workload that is not business-critical or too difficult to migrate, has minimal dependencies on other environments, yet is typical enough to serve as a blueprint for later waves. Migration Center helps inventory existing workloads, while the fit assessment tool helps determine if a virtual machine workload is fit for modernization to a container or for migration to Compute Engine.
KPIs and metrics provide the shared language for data-driven decisions about architecture, capacity, and optimization. Key performance indicators (KPIs) are quantifiable measurements that reflect overall performance, while metrics are specific data points that contribute to the calculation of KPIs. A KPI such as application health or cost efficiency is calculated from underlying metrics, and regular review of those metrics reveals trends, patterns, and anomalies that are not visible without systematic measurement. Cloud Monitoring enables real-time monitoring of cloud resources and services, and BigQuery stores and analyzes the data gathered through monitoring. Used together, Cloud Monitoring and BigQuery feed custom operational dashboards that provide a centralized view of the most important metrics. A dashboard might include CPU utilization, memory usage, network traffic, and latency for a particular application, which lets teams quickly identify areas that need attention and take steps to resolve them.
For hybrid and multicloud decisions, success criteria must be measurable with clear milestones for each migration phase or wave. Service-level agreements (SLAs) and KPIs must be defined for applications that have distributed components across multiple environments, and they must be validated with both functional and load testing before moving to production. Capacity planning also depends on this data, using historical usage patterns, growth projections for users and features, business requirements like SLAs and performance targets, and allowances for seasonal spikes, product launches, and graceful failover between primary and disaster recovery (DR) systems.
Cost and value analysis turns raw cloud spending data into contextualized knowledge that justifies infrastructure investment. Cloud Billing reports and the FinOps Hub (opens in a new tab) provide comprehensive cost monitoring for finance and audit teams, while billing data exported to BigQuery and visualized in Looker or custom dashboards contextualizes spend against business units, profitability, and regulatory reporting. Tagging and labeling resources by project, team, or department, supported by cost allocation and chargeback, attaches each cost to its owner and creates accountability for responsible resource management. The data-information-knowledge-wisdom (DIKW) pyramid describes how raw spend becomes a decision in order:
app1-test-vmA running 730 hours in us-central1 for USD 70.n4-standard-2 VMs across two regions.app1 testing team spending USD 650 during stress testing in the second week of July.Cost-benefit choices are then evaluated with budgets, alerts, and pricing options rather than service price alone. Cloud Billing budgets and alerts set spending limits per team or project, provide real-time insight into actual versus budgeted spend, and notify stakeholders of potential overruns so corrective action happens before significant deviation. Google Cloud cost optimization options include:
Active Assist systematically identifies idle resources and underutilized commitments across Compute Engine, Cloud Run, BigQuery, and Cloud SQL without configuration. Regular cost postmortems, proactive cost reviews, and reviews of pricing models ensure forecasts use up-to-date tags, labels, and features, and that short-term savings are weighed against long-term value, risk mitigation, and total cost of ownership.
Exam tip: Metrics are the raw data points like CPU or latency, while a KPI is the quantifiable measurement of overall performance calculated from those metrics.
Architectural choices are approved on business drivers, while technical drivers explain the gain or constraint that supports or threatens those business goals. Business effects are evaluated by potential for market differentiation through capabilities like artificial intelligence on on-premises data, potential savings in total cost of ownership, potential improvements in availability, resiliency, security, or performance such as adding a cloud DR site, and potential speedup of development and release through cloud development and testing environments. Drivers must be considered together as a balance of priorities, not in isolation, because a single-cloud approach can reduce complexity and offer built-in integrations and CUDs, while multicloud can add management, security, integration, performance, skills, and pricing-tool costs. Technical drivers must be translated into business impact to justify hybrid or multicloud patterns. Common technical drivers include building analytics and AI capabilities that are difficult on premises, improving quality and performance of service, automating rollouts for faster time to market, using high-level APIs to speed development, accelerating provisioning of compute and storage, using serverless services for elastic scale, and using global infrastructure for multi-regional architectures. The most common temporary driver is facilitating migration itself, where enterprises transition applications and data in prioritized migration waves and use on-demand pricing and high compute capacity for a limited research, scale, merger-integration, or proof-of-concept need.
Governance frameworks and stakeholder alignment are the structures that connect cloud architecture decisions to regulatory, financial, and operational requirements. Governance frameworks define the policies, controls, and evaluation processes that a cloud initiative must satisfy, while stakeholder alignment assigns clear ownership for choosing, approving, and enforcing those controls. Together they ensure that technical choices about identity, data, infrastructure, and operations remain consistent with enterprise goals.
Stakeholder roles in the decision-making lifecycle define who authorizes, implements, and reviews cloud controls throughout the system life cycle. The customer is the expert on business, regulatory, and risk requirements and must identify the security controls each workload needs based on compliance obligations, organizational standards, and customer and vendor requirements. Google provides the foundational infrastructure, default encryption, and infrastructure controls, and enforces the customer-defined policies for workload administration and data access. A configuration management plan formalizes these roles by addressing responsibilities, processes for identifying configuration items, and procedures for managing those items from development through operations. The plan must be reviewed and approved by defined personnel or roles and protected from unauthorized disclosure and modification. Management also carries ongoing responsibility to balance ongoing and separate evaluations, build evaluations into business processes, and select control activities that ensure completeness, accuracy, and availability of technology processing.
Customer responsibilities in the shared model include:
Google responsibilities in the shared model include applying default encryption and infrastructure controls, enforcing customer-set IAM policies, and providing Organization Policy Service (opens in a new tab), resource settings, and Policy Intelligence tools that give insight into access to accounts and resources. For regulated workloads, Sovereign Controls by Partners (opens in a new tab) provides additional configurations and controls for data residency, storage location, and limits on personnel access for protected data types in configured resources.
The shared responsibility model divides security ownership between Google and the customer, and the division shifts with the service model. Google always remains responsible for the underlying network and infrastructure, while the customer always remains responsible for access policies and data. The choice of service model therefore determines how much application-level and platform control the customer retains.
| Service model | Examples | Responsibility split |
|---|---|---|
| Infrastructure as a service (IaaS) | Compute Engine virtual machines, networks, storage | Customer holds most security responsibilities; Google is responsible for underlying infrastructure and physical security |
| Platform as a service (PaaS) | App Engine, Google Kubernetes Engine (GKE) (opens in a new tab), BigQuery | Google owns more controls depending on features used; responsibility is shared for application-level controls and IAM management, while customer remains responsible for data security and client protection |
| Software as a service (SaaS) | Google Workspace, Google Security Operations, Marketplace applications | Google owns most security responsibilities; customer remains responsible for access controls and data stored in the application |
| Function as a service (FaaS) or serverless | Cloud Run functions that run single-purpose functions in response to events | Similar to SaaS; customer defines the triggering event and code behavior, Google manages the platform and scaling |
Exam tip: Google always retains the underlying network and infrastructure, while the customer always retains access policies and data, regardless of IaaS, PaaS, SaaS, or FaaS.
Governance frameworks for regulatory and operational requirements are published control sets that define what must be in place and what evidence auditors expect. Common frameworks mapped to Google Cloud controls include NIST SP 800-53 Rev. 5 (opens in a new tab), SOC 2, ISO/IEC 27001:2022, Payment Card Industry Data Security Standard (PCI DSS), Health Insurance Portability and Accountability Act (HIPAA), General Data Protection Regulation (GDPR) assurances, and NIST AI 600-1 privacy controls for generative AI. Inherited controls such as default encryption and infrastructure design can be used as evidence of security posture, while customer-configured controls must be implemented and monitored by the customer. The frameworks converge on the same governance themes: restrict to least functionality, manage secure configurations, control identity and access, protect data, and monitor continuously. Secure configuration means establishing the most restrictive mode consistent with operations, implementing it, approving deviations, and monitoring changes, including blocking serial ports, disabling legacy networks and legacy metadata endpoints, and enabling DNSSEC for Cloud DNS. Access governance means provisioning, reviewing, modifying, and removing access rights under policy, including blocking public IP addresses for AlloyDB and Cloud SQL, enforcing public access prevention for Cloud Storage, restricting public access to BigQuery datasets, and controlling service account keys and API key restrictions.
Operational governance requires detection, availability, and retention controls that flow from monitoring sources to responsible reviewers. Detection controls send data from VPC Flow Logs, subnet flow logs, firewall rule logging, Cloud DNS logging, Cloud Logging and Cloud Monitoring on GKE, and audit logs to analysis and alerting, with detected events communicated to defined essential contacts. Availability controls depend on automated recovery and protection, including auto repair for GKE clusters, automatic backups for Cloud SQL, enforcement of SSL for database connections, and restriction of legacy TLS versions. Retention governance depends on defined lifecycle rules, where Admin Activity and System Event audit logs are retained for 400 days and Data Access audit logs for 30 days by default before data flows to customer-selected storage for longer retention.
Organizational alignment through hierarchy and policy is how stakeholder decisions are enforced consistently across folders, projects, and applications. The resource hierarchy of organization, folder, and project creates the trust and management boundary where policies apply, so sensitive data with residency or access requirements stays inside compliant Sovereign Controls by Partners folders or projects. Organization Policy Service constraints and resource settings then propagate those decisions downward, while Policy Intelligence tools analyze which resources are covered by which policy. An organizational security mindset sustains this alignment by treating security as part of system design, assuming zero trust, integrating security through development, and planning for cyber defense, responsible AI use, and regulatory and privacy needs. Focus areas divide the work so owners are clear: infrastructure security secures networks, encrypts data in transit and at rest, and controls traffic flow with Cloud Next Generation Firewall, VPC Service Controls, and Google Cloud Armor; identity and access management manages authentication, authorization, and policies with Cloud Identity, IAM, and Workforce and Workload Identity Federation; and data security stores, classifies, and protects data with encryption, access controls, and loss prevention. Sensitive or regulated data in scope for these decisions includes account information, health information, personal identifiers, cardholder data, and ID numbers, with additional residency and security controls available for Europe and the EU.
Trade-off analysis and technical debt management is the practice of weighing rapid deployment against long-term maintainability when designing cloud architecture. Every decision to add availability, expand to another region, or build custom logic creates future cost in operations, complexity, and risk. Technical debt accumulates when short-term shortcuts for speed leave manual processes, single points of failure, or locked-in designs that are expensive to change later.
The core architectural trade-off balances availability against operational complexity, latency, and cost. Increasing the number of nines of availability generally costs more in redundant resources, cross-region transfer, and operational effort, but higher availability helps meet regulatory requirements for critical workloads. For example, 99.9% availability allows about 86 seconds of downtime in a 24-hour day, while 99% allows 864 seconds, which is 10 times more downtime. Architects must identify required availability, accurately calculate its impact, and assess how each design choice affects both architecture and budget.
Aggregate availability is the overall uptime of a multi-tier system, calculated from the availability of each tier or component in the stack. For chained services where each tier depends on the next, aggregate availability is the product of each tier: tier1_availability × tier2_availability × tierN_availability. Four tiers at 99.9% each yield only 99.6% aggregate availability, and two services at 99% chained together yield only 98%. Where feasible, architects should choose parallelization over chaining, because parallelized services run independently and use the formula 1 − (1 − A) × (1 − B), so two 99% services in parallel yield 99.99%.
Exam tip: High availability (HA) applies to regional deployments with synchronous replication between zones, while disaster recovery (DR) applies to multi-region deployments with typically asynchronous replication.
High availability (HA) and disaster recovery (DR) solve different failure scopes with different replication mechanisms. Spanner (opens in a new tab) and Cloud Storage (opens in a new tab) are examples of managed multi-regional services that support active-active topologies without custom replication logic.
| Option | Scope | Replication | Recovery objectives and failover |
|---|---|---|---|
| High availability (HA) | Single region across zones | Synchronous replication between zones by default | Zero or near-zero recovery time objective (RTO) and recovery point objective (RPO) with active-active topology and no single point of failure (SPOF) |
| Disaster recovery (DR) | Two or more regions | Typically asynchronous unless using multi-regional or global services | Higher RTO and RPO with manual or semi-automated failover operations that must be defined and tested |
DR planning requires well-defined plans for zonal and regional outages, including data replication for regional or zonal resources, snapshots of persistent disks stored in multi-region locations, and regular testing of failover. For regulated workloads, the choice of failover region may be limited by data sovereignty and data residency rules, and RTO and RPO must align with the impact tolerance required by financial regulations.
Managed services trade some custom control for built-in backups, HA backed by uptime service-level agreements (SLAs), scalability, and resilience features that reduce long-term maintenance. For data management, Cloud SQL, Cloud Storage, and Spanner provide HA and replication without custom clustering code. For compute, Compute Engine (opens in a new tab) managed instance groups (MIGs) and Google Kubernetes Engine (GKE) (opens in a new tab) regional clusters automatically recreate failed virtual machines (VMs) and survive zone outages, while managed multi-regional services improve resilience against region outages. To avoid creating new technical debt through lock-in, architects must define exit plans for services with unique characteristics and verify portability before signing contracts:
Automation reduces technical debt by replacing manual provisioning and recovery steps with consistent, repeatable processes that lower human error and speed recovery. Infrastructure as code (IaC) tools like Terraform define infrastructure once so environments are recreated the same way every time. Event-driven remediation flows from audit logs through Eventarc (opens in a new tab) or Workflows to corrective actions, while autoscaling adds capacity automatically during failover and platform engineering applies regulatory policies and guardrails during deployment. Operability debt also grows with deployment footprint, so architects should eliminate single points of failure and match the archetype to the need. A zonal deployment is simplest to operate, while redundant multi-zone or multi-region stacks require monitoring of each stack and component, graceful handling of in-process requests, careful rollouts, and database synchronization across locations. Global resources like a global load balancer can lower setup effort compared with managing multiple regional resources but require careful change management and Premium Tier networking, while hybrid and multicloud topologies add networking, consistent management, and nonstandard security tooling costs that must be weighed per unit of work such as users served or requests processed.
Metrics are specific data points such as CPU utilization or latency that contribute to the calculation of key performance indicators (KPIs), which are quantifiable measurements of overall performance such as application health or cost efficiency. Cloud Monitoring enables real-time monitoring of cloud resources and services and BigQuery stores and analyzes the data, and together they feed custom operational dashboards that provide a centralized view of the most important metrics.
For chained services where each tier depends on the next, aggregate availability is the product of each tier availability, so four tiers at 99.9 percent each yield only 99.6 percent and two services at 99 percent chained together yield only 98 percent. Where feasible architects should choose parallelization over chaining because parallelized services use 1 minus the product of their unavailabilities, so two 99 percent services in parallel yield 99.99 percent.
High availability (HA) covers a single region across zones with synchronous replication between zones by default, while disaster recovery (DR) covers two or more regions with typically asynchronous replication. High availability (HA) provides zero or near-zero recovery time objective (RTO) and recovery point objective (RPO) with active-active topology and no single point of failure, while disaster recovery (DR) has higher recovery time objective (RTO) and recovery point objective (RPO) with manual or semi-automated failover operations that must be defined and tested.
Migration risk assessment evaluates potential effect of migration-caused outages on business operations, experience of the team with public cloud deployments or a new or second provider, and need to comply with legal or regulatory restrictions such as data sovereignty. Technical difficulty focuses on how hard the workload will be to move and run, including size, complexity and age of the application and infrastructure, number of dependencies with other applications and services, and restrictions imposed by third-party licenses and specific versions of operating systems, databases or other configurations.
An online retail enterprise is evaluating whether to modernize its on-premises monolithic checkout application to a multi-region deployment on Google Cloud. Executive leadership requires a data-driven decision framework that balances business risks, infrastructure costs, and technical performance before approving the migration budget.
The application experiences severe performance bottlenecks and occasional downtime during seasonal sales events, directly resulting in lost revenue and degraded customer satisfaction.
Which strategy should the Cloud Architect recommend to assess the risks and justify the architectural investment?
Last updated