Prepare and test your skills
Prepare and test your skills
Definition of Quality Attributes and Performance Metrics is the practice of turning business needs for availability, scalability, and performance into measurable non-functional requirements. Quality attributes describe how well a system must behave during normal load and failure, while performance metrics quantify that behavior over time. In Google Cloud, a service level indicator (SLI) provides the measurement and a Service Level Objective (SLO) (opens in a new tab) sets the target the system must meet.
Quality attributes for cloud solutions are the non-functional requirements that keep a global system resilient and responsive. Availability requires that platform services continue without interruption when hardware, software, or network fails, while scalability requires that the system grow and shrink automatically during traffic spikes. Performance requires a fast and consistent user experience supported by low latency and sustained throughput. Together these three attributes determine technical thresholds such as uptime percentages, percentile latencies, and recovery time limits.
Google Cloud supports these attributes with highly redundant infrastructure built on redundancy of everything. Redundancy applies across peer areas:
Because data is mirrored automatically across storage devices in multiple locations, the system does not depend on a single server, data center, or network connection. Customers extend the same model by deploying resources across multiple regions and zones, and managed services such as Cloud Run, Compute Engine, and Firestore provide automatic scaling as demand changes.
A service level indicator (SLI) is a quantitative measure of a specific aspect of service behavior seen from the user's perspective. The two core SLIs are latency and availability. Latency measures how long it takes to return a response to a request, usually in milliseconds, and raw data collected over time is calculated as percentiles to show variation. Availability measures the fraction of time a service responds successfully, presented as the ratio of successful responses to total responses.
In Cloud Service Mesh (opens in a new tab), SLI data flows from service telemetry to the Metrics page for each service. The Latency graph shows latency over time and helps determine a latency threshold or upper bound, while the Error rate graph helps determine availability. Broader observability uses the four golden signals to capture health in the same way:
Saturation shows how close the system is to capacity limits, so traffic growth can be linked to latency increases and error-rate changes before users are affected.
A Service Level Objective (SLO) is a target value for a service level measured by an SLI over a period of time. An SLO is expressed as SLI ≤ upper_bound or SLI ≥ lower_bound, which turns a measurement into a clear pass-or-fail goal. SLOs give a concise and low-noise signal of overall health, so teams can monitor one objective instead of reacting to every short spike in raw metrics.
SLOs are set and viewed on the Health page based on telemetry data, and alerts are created in Cloud Monitoring to warn when a service is not performing as expected. Concrete SLO targets include 99.9% of API calls must return a successful response, 95th percentile inference latency must be below 300 ms, time to first token (TTFT) must be below 500 ms for 99% of requests, and rate of harmful output must be below 0.1%. Two general examples show the same pattern with time windows: latency can exceed 300 ms in only 5 percent of requests over a rolling 30-day period, and the system must have 99% availability measured over a calendar week.
Exam tip: An SLI is the measurement, such as percentile latency or successful-response ratio, while an SLO is the target with a bound and time window, such as below 300 ms for 95% of requests over 30 days.
Technical thresholds for global resilience and low latency translate SLOs into design choices for placement, routing, and recovery. For resilience, recovery is judged against a Recovery Time Objective (RTO) for how long resumption may take and a Recovery Point Objective (RPO) for how much data loss is acceptable. For performance, thresholds use percentile bounds such as 95th percentile latency and TTFT, plus success rate and throughput, to protect low-latency user experience during normal load and failover.
Disaster recovery (DR) (opens in a new tab) patterns select the level of standby readiness needed to meet RTO and RPO values:
A three-column comparison of Cold, Warm, and Hot DR patterns by standby readiness, recovery action, and outage impact from largest to smallest RTO.
| Pattern | Standby readiness | Behavior during failure |
|---|---|---|
| Cold | No running spare capacity; recovery resources must be provisioned | Service stops until help arrives and repair completes, which suits compliance data with large RTO |
| Warm | Spare capacity and replacement kit are prepared but stopped | Service pauses while traffic shifts to prepared capacity, which suits non-critical systems such as email notification |
| Hot | Fully running redundant capacity across zones and regions | Service continues with little immediate impact, which suits online user-facing services that need fast recovery |
Google's global backbone network supports these thresholds with software-defined networking, edge caching, and multiple points of presence (PoPs) that route traffic over Google's backbone rather than the public internet. When a zone or region disconnects, local services in the affected area may go offline as expected while control planes and services in other zones and regions remain unaffected. Load balancers and health checks then reroute traffic from the failed zone or region to healthy zones or regions, so latency, error rate, and throughput stay within SLO bounds.
Monitoring health and verifying recovery confirm that quality attributes and SLOs hold during and after a failure simulation. Telemetry flows from load balancers, health checks, and application services to Cloud Monitoring and Cloud Logging, where metrics and events are captured during the test. Teams observe changes in latency, error rates, and throughput, and they verify that logs are generated and alerts are triggered for key events such as service outages or failovers.
Recovery is verified by measuring how long the system takes to resume normal operations and comparing the result with the defined RTO, then documenting any gaps. Data integrity and availability are checked against the RPO by comparing database snapshots or backups before and after the failure, and service restoration is confirmed with minimal user disruption. Each test step, failure scenario, timestamp, log, and metric is documented to highlight bottlenecks, single points of failure, or unexpected behavior, and failover policies and playbooks are updated from the findings.
Translation of business processes into functional requirements means converting a business use case into specific statements about what the system must do, where it must run, and how data must flow. The business goal of listing and selling a product on Google Cloud Marketplace (opens in a new tab) becomes functional requirements for organization setup, product readiness, hosting location, data movement, and service integration. Each requirement names the responsible owner and the Google Cloud environment that must support increased customer consumption.
Organizational prerequisites are the functional entry conditions that allow a partner organization to list, sell, and receive payment through Google Cloud Marketplace. The organization owns compliance with these conditions, and Google Cloud depends on them to create the vendor account and payment flow. Without these conditions, the listing workflow cannot proceed to product validation.
Product readiness requirements define the application logic baseline that a product must satisfy before it can be publicly listed and sold. The product must be production-ready and must not be in an alpha or beta state. The product must be enterprise-ready with a professional online presence, a defined sales motion, customer support, and strong security practices, and it must contain no known vulnerabilities, viruses, spyware, Trojan horses, or other malicious code.
The product must host its software primarily on Google Cloud, as verified through an approval process during onboarding. Products that meet all listing requirements and pass a business case review during internal validation receive a standard revenue share, while products that do not meet requirements or do not pass review may still be offered with an adjusted revenue share. If changes to the product or organization affect compliance or invalidate onboarding documentation, the partner must notify Google and submit the product for re-review and re-approval.
Hosting patterns for compute and AI agents specify where the compute plane, data plane, and control plane run and which side of the trust boundary carries growth. These patterns translate the business architecture into service integration requirements for Google Cloud, on-premises environments, and other clouds.
Exam tip: Pattern 2 and Pattern 9 share the same decision rule — the Google Cloud-hosted compute, data plane, or agent must grow faster than any off-cloud component when usage grows.
Hosting patterns for data and deployment specify the source and destination of data flows and where storage, analysis, and execution must occur. In these patterns, data flows from agents, source systems, or edge devices into a Google Cloud-hosted environment for storage, analysis, delivery, or migration.
Additional product requirements for parity, privacy, and consumption ensure functional parity, data protection, and measurable Google Cloud consumption after listing. The product on Cloud Marketplace must have the same capabilities and features as any version offered outside Cloud Marketplace. Data products must not contain personally identifiable sensitive information as defined in the Protecting Americans' Data from Foreign Adversaries Act of 2024.
Products with components deployed into customer-owned Google Cloud projects must implement customer tenant consumption tracking for Software as a Service (SaaS), VM, or Kubernetes solutions, so usage flows from the customer tenant to the tracking system. Every listing must result in meaningful Google Cloud consumption by the customer procuring or using the solution, which connects the business use case directly to resource usage on Google Cloud.
Assessment of security, compliance, and operational constraints is the process of turning business rules, laws, and limits into non-functional requirements for a Google Cloud design. These requirements define where data must live, how it must be protected, how much it can cost, and who is responsible for each control. A design that meets these constraints stays reliable, scalable, and secure while satisfying auditors, regulators, and corporate governance.
The shared responsibility model (opens in a new tab) divides security work between Google Cloud and the customer across all service types. Google Cloud always remains responsible for the underlying network and physical infrastructure, while the customer always remains responsible for access policies and data. In software as a service (SaaS) and function as a service (FaaS), including Cloud Run functions, the customer remains responsible for access controls and for the data chosen to store in the application.
This division means an architect must identify which controls are customer-owned, which are available as part of the cloud offering, and which are inherited by default. Inherited controls, such as default encryption and infrastructure controls, can be used as evidence of security posture to auditors and regulators. A common operational challenge is misconfiguration, because cloud products change often and new products launch constantly, and many breaches result directly from that misconfiguration.
Industry and regulatory frameworks define mandatory security controls that become non-functional requirements for workloads moved to Google Cloud. The Payment Card Industry Data Security Standard (PCI DSS) defines regulations for payment processors, and responsibilities are shared between the customer and Google Cloud as described in the Google Cloud PCI DSS Shared Responsibility Matrix. The Health Insurance Portability and Accountability Act (HIPAA) sets standards for handling electronic personal health information (PHI), with shared duties described in Google Cloud HIPAA compliance guidance.
Other industries, such as finance and manufacturing, have their own rules for how data can be gathered, processed, and stored, with details in the Compliance resource center. For U.S. federal workloads, the Federal Risk and Authorization Management Program (FedRAMP) (opens in a new tab) requires a Rev. 5 Agency Authorization to Operate (ATO) completed with Google and a third-party assessment organization (3PAO). Google supports this path with a Customer Responsibility Matrix (CRM) that details customer duties for NIST SP 800-53 controls in the FedRAMP High baseline, a System Security Plan (SSP) that describes the authorization boundary and control implementation, plus service-specific control mapping guides and FedRAMP High implementation guides for services such as BigQuery, Compute Engine, Cloud Storage, Cloud Key Management Service, Google Kubernetes Engine (GKE), and Identity and Access Management (IAM).
| Authorization level | Control baseline | Data boundary requirement |
|---|---|---|
| FedRAMP High | Full FedRAMP High control baseline, including data residency control SA-9(5) | Must use Assured Workloads Data Boundary for FedRAMP High |
| FedRAMP Moderate | Subset of the High baseline with fewer controls to assess | Can use any FedRAMP High authorized service inside the Moderate authorization boundary |
Location-based constraints turn the location of offices, customers, and data into data residency requirements. Different countries and regions regulate how customer data can be processed and stored, so the architect must map each dataset to an allowed Google Cloud region. For example, if customers reside in the European Union, the business may need to follow the General Data Protection Regulation (GDPR) and keep customer data in Google Cloud regions in the EU.
Data residency enforcement in Google Cloud depends on the compliance baseline chosen for the workload. FedRAMP data residency requirements are mandated only by the FedRAMP High control baseline, and they are enforced by choosing the Assured Workloads Data Boundary for FedRAMP High. The customer is responsible for ensuring collected data remains in the selected boundary and regions, while Google provides GDPR and regional compliance offerings to show how platform duties are met.
Exam tip: FedRAMP High mandates data residency through the Assured Workloads Data Boundary for FedRAMP High, while FedRAMP Moderate does not add a separate residency mandate but can inherit FedRAMP High authorized services.
Encryption and access requirements define how data is protected at rest, in transit, and during access decisions. Inherited platform protections include default encryption and infrastructure controls, which reduce the controls a customer must build from scratch. Inside Google infrastructure, remote procedure call (RPC) traffic between services is secured with Application Layer Transport Security (ALTS), and each server holds a unique machine identity tied to hardware roots of trust that can be revoked and rotated after a security incident.
Customer-owned non-functional requirements focus on identity, context, code integrity, and auditability. Strong authentication requires cryptographic credentials for jobs and two-factor methods for users, while context-aware access evaluates user identity, location, device ownership, configuration, and fine-grained policy instead of trusting the corporate network alone. Production software must be reviewed and approved before deployment through controls such as Binary Authorization for Borg, which alerts owners when code and configuration do not meet standards.
Logging and monitoring complete the control set by making access visible and auditable:
Incident detection combines these logs with threat intelligence, analysis of past incidents, network traffic, system access logs, anomalous behavior, and offensive security exercises.
Budgetary constraints turn funding limits and business goals into cost and performance requirements. Technology initiatives draw from different budgets, with innovation and transformation efforts sometimes called change-the-bank and business-as-usual operations called run-the-bank, each with different access to resources. To align spending with value, each initiative should connect to business objectives and key results (OKRs) such as increasing revenue or profit, reducing costs, and mitigating risk more efficiently, and optimization should focus at the system level rather than only within front-office versus back-office silos.
Security and compliance standards must remain high without rejecting valuable technology based on unproven risk assumptions. Architects should confirm with Google Cloud account, risk, and product teams that intended products can meet security, resilience, and compliance needs for each operating jurisdiction. For data-intensive and AI workloads, transparency supports both performance and audit needs through AI explainability, citations to sources, grounding, and retrieval-augmented generation (RAG), and when explainability is not enough, the design should separate decision-making steps from non-decision steps and keep human intervention where regulations such as GDPR Article 22 require it.
A service level indicator (SLI) is a quantitative measure of service behavior seen from the user's perspective, such as percentile latency or successful-response ratio, while a Service Level Objective (SLO) is the target value for that SLI over a period of time. An SLO is expressed as an SLI less than or equal to an upper bound or greater than or equal to a lower bound, for example 99.9 percent of API calls must return a successful response and 95th percentile inference latency must be below 300 milliseconds.
Cold, warm, and hot disaster recovery (DR) patterns provide no running spare capacity, prepared but stopped capacity, and fully running redundant capacity in that order of faster recovery. Cold suits compliance data with large Recovery Time Objective (RTO) values where service stops until repair completes, warm suits non-critical systems such as email notification where service pauses while traffic shifts to prepared capacity, and hot suits online user-facing services that need fast recovery where service continues with little immediate impact.
A product listed on Google Cloud Marketplace must be production-ready and not in an alpha or beta state, must host its software primarily on Google Cloud, and must drive meaningful Google Cloud consumption by the customer procuring or using the solution. The partner organization must maintain good standing in Google Cloud Partner Network (Partner Network), be incorporated in a supported region for receiving payments, and maintain a Cloud Marketplace vendor account and payment profile in good standing.
Federal Risk and Authorization Management Program (FedRAMP) High requires the full High control baseline including data residency control SA-9(5) enforced with the Assured Workloads Data Boundary for FedRAMP High. Federal Risk and Authorization Management Program (FedRAMP) Moderate uses a subset of the High baseline with fewer controls to assess, does not add a separate residency mandate, and can use any FedRAMP High authorized service inside the Moderate authorization boundary.
A financial enterprise is designing a centralized analytics platform on Google Cloud using BigQuery to process sensitive customer financial records. The solution must adhere to strict regulatory, compliance, and governance non-functional requirements:
europe-west3 (Frankfurt) region.Which architectural combination should you implement to satisfy these compliance requirements?
Last updated