NVIDIA NCA-AIIO Exam Prep
AI Infrastructure and Operations (Page 11 )

Question 3: AWS Storage Gateway File Gateway

• A: S3 bucket per user — S3 is object storage, not a traditional shared file system; mounting it requires extra tools and is less efficient.
• B: File Gateway — Provides a local file share using standard file protocols while storing files in Amazon S3. This preserves convenient local access and extends storage capacity.
• C: WorkSpaces and WorkDocs — Designed for virtual desktops and document collaboration, not large shared file storage.
• D: EC2 with EBS — EBS is attached to one EC2 instance and is not designed for direct multi-user file sharing.

• Leaf switches provide access ports for servers, users, and other devices.
• Spine switches provide the backbone between leaf switches.
• Every leaf switch connects to every spine switch, creating a predictable, full-mesh fabric between the two layers.
• Leaf switches do not connect directly to other leaf switches, and spine switches do not connect directly to other spine switches.

• A: Adding both a spine and leaf is unnecessary for simply adding access ports.
• B: It focuses on uplink speed, not the scalable topology.
• D: A single spine connection creates a bottleneck and a single point of failure.

• Install unauthorized or unapproved applications
• Access system areas normally restricted
• Add features unavailable in the default software
• Modify the device’s operating system

• SQL injection (SQLi): Attacks databases by inserting malicious SQL commands.
• Cross-site scripting (XSS): Injects malicious scripts into web pages viewed by users.
• Sideloading: Installs an application from outside the official app store, but it does not necessarily remove the phone’s operating-system restrictions.

• Laws and regulations
• Litigation holds
• Industry or contractual obligations
• Tax, financial, health, or privacy requirements

• Business strategy and direction: May influence retention, but cannot override legal requirements.
• Storage capacity and longevity: Practical considerations, not the primary driver.
• Business ease and value analysis: May support decisions after compliance obligations are identified.

• Sensitive document shared during a Teams chat: Teams chat
• Sensitive document stored/shared in a Teams channel: Teams channel messages

• Files shared in a chat are generally stored in the sender’s OneDrive, but the relevant DLP context is the Teams chat message and its attachment.
• Files shared in a channel are stored in the team’s SharePoint site, but the relevant context is the Teams channel message.
• The DLP rule can then use an action such as restrict access to the content to prevent guest users from opening the sensitive document.

• B. Deploy to a SageMaker endpoint: This is needed for SageMaker-hosted models, not for a model intended to be used directly in Amazon Bedrock.
• C. Register with SageMaker Model Registry: SageMaker Model Registry is unrelated to invoking custom models in Bedrock.
• D. Grant access in Bedrock: Access permissions may be necessary, but they do not replace the required provisioned capacity.

• The application failing to start or becoming unhealthy
• An incorrect probe path, port, or protocol
• The application needing more startup time
• Dependency or configuration failures
• A liveness probe being too aggressive

bash 
kubectl describe pod <pod-name> 
kubectl logs <pod-name> --previous

• Scale to more replicas: may hide the issue but does not diagnose it.
• Decrease initialDelaySeconds: makes probes run earlier and could worsen startup failures.
• Drain and reboot the node: node metrics are normal, so rebooting is not justified as the first step.

• Reviewing security policies and standards checks whether controls are documented, not whether they work.
• Reviewing security logs may show that monitoring is occurring, but does not directly establish that preventive controls are effective.
• Analyzing technical security practices is broad and may assess implementation, but it is less objective than an actual controlled attack.

• Design/compliance effectiveness: policies, standards, and procedures
• Operational effectiveness: logs, records, and evidence that controls operate
• Technical security effectiveness: penetration testing or vulnerability testing

• C: Develop the rule in the Rules Editor, define the events, match, and condition sections, then use the test rule feature. This directly validates the YARA-L rule without enabling it in production.
• A: UDM Search can help investigate data and refine filters, but it does not fully test YARA-L logic—especially complex event joins, match windows, and conditions.

1. You try to access a cloud application, such as accounting software.
2. The application redirects you to the company’s IdP.
3. You authenticate with the IdP.
4. The IdP sends a digitally signed SAML assertion to the application.
5. The application validates the assertion and grants access.

• The user’s identity
• Whether authentication succeeded
• Roles or group membership
• Access-related attributes

• Managerial: policies, standards, risk management, governance
• Operational: procedures carried out by people, such as training, incident response, and backups
• Technical: firewalls, encryption, access-control systems
• Physical: locks, guards, cameras

• Red team: Simulates attackers; exploits weaknesses to test security.
• Blue team: Defends systems; monitors, detects, and responds to attacks.
• Purple team: Combines red- and blue-team efforts to improve detection and defense.
• White team: Oversees or manages the test, defining rules, scope, and authorization.

• Approve major information security policies
• Allocate funding and resources
• Set the organization’s risk appetite
• Ensure security supports business objectives

• Evaluating vendors is generally handled by security, procurement, or technical teams.
• Assessing risks is typically performed by security and risk-management professionals, with management accepting or prioritizing the risks.
• Monitoring regulatory adherence is usually handled by compliance, audit, or security functions.

• Asynchronous inference is designed for applications that submit individual inference requests, place them in a queue, and retrieve results later. It is useful when requests may take a long time or when traffic is intermittent.
• Batch transform is designed specifically to run inference over an entire large dataset stored in Amazon S3 and produce an output dataset.

• Large dataset/offline processing ? Batch transform
• Large or slow individual requests processed later ? Asynchronous inference

1. Select Exchange email as a location.
2. Use the built-in Azure Storage Account key sensitive information type as the condition.
3. Configure the DLP action to encrypt the email—for example, using Microsoft Purview Message Encryption.

• Reviewing security objectives to ensure they support business goals.
• Ensuring security requirements are integrated consistently across business units.
• Coordinating security policies, risk management, controls, and awareness activities.

• Executive management: Provides direction, funding, and support, but usually does not manage day-to-day security integration.
• Board of directors: Provides governance and oversight, not operational security coordination.
• Steering committee: Helps coordinate and prioritize initiatives, but the CISO typically owns the overall security program and ensures its integration.

• Tokenization: Replaces the value with a nonsensitive token, often reversible through a secure token vault.
• Hashing: Converts data into a one-way fixed-length value, generally used for integrity checks or password storage.
• Obfuscation: A broad term for making data difficult to understand; it is less precise than masking in this context.

• Honeypots: Deliberately vulnerable or attractive systems designed to lure attackers. Legitimate users should not access them, so activity on a honeypot is a strong indicator of compromise.
• DNS sinkholes: Redirect requests for known malicious domains to a controlled internal system instead of allowing the connection. This can identify infected hosts attempting to contact command-and-control infrastructure.

• Tokenization: Replaces sensitive data with non-sensitive tokens; it protects data but does not primarily detect attackers.
• CI/CD: Automates software development and deployment processes.
• Threat modeling: Identifies potential threats and weaknesses before or during system design.
• Data obfuscation: Makes data harder to understand, but it is not an attacker-detection method.

• CSIRT: A Computer Security Incident Response Team that handles and coordinates incident response.
• CVSS: A scoring system used to rate the severity of vulnerabilities.
• SOAR: Security Orchestration, Automation, and Response technology used to automate security workflows.

• Third-party enterprise application with embedded AI: The provider manages most security controls. The customer mainly manages users, data, and configuration.
• Application using an existing third-party FM: The customer controls the application and its inputs/outputs, but the provider controls the underlying model and infrastructure.
• Fine-tuning a third-party FM: The customer has additional responsibility for training data, data protection, and model behavior, but still depends on the provider’s base model and platform.
• Building and training a model from scratch: The customer controls the model, training data, application, deployment, and many security safeguards. This creates the greatest security ownership.

xml 
<apex:page> 
    <apex:stylesheet value="{!URLFOR($Resource.MyStyles, 'styles/main.css')}" /> 
</apex:page>

• <apex:includeScript> is for JavaScript.
• <apex:includeStyle> is not a valid Visualforce component.
• <apex:require> is not the Visualforce tag used to load CSS.

text 
ip route vrf Customer1 172.16.1.0 255.255.255.0 172.16.1.1 global 
ip route vrf Customer2 192.168.1.200 255.255.255.255 192.168.1.1 global 
ip route 192.168.1.0 255.255.255.0 Vlan10 
ip route 172.16.1.0 255.255.255.0 Vlan20

• The route in Customer1 VRF provides a path back to the Customer2 subnet, 172.16.1.0/24.
• The route in Customer2 VRF provides a path to the FTP server, 192.168.1.200.
• The keyword global tells the router to resolve the next hop using the global routing table, enabling controlled route leaking between the VRF and global table.
• The global routes identify the VLAN interfaces used to reach each customer network.

• An Adj-SID does not necessarily need manual configuration.
• It can often be dynamically allocated by the router, although manual configuration may also be possible.
• The adjacency information is advertised through the IGP, such as IS-IS or OSPF.

• It is identified by an Adj-SID.
• It represents a particular adjacency of a node.
• Its meaning is generally local to the advertising node; unlike a Node-SID, it is not a globally reusable instruction across the network.

• Database data: Aggregates are used to retrieve structured data, commonly from entities and database sources.
• Screen scope: A Screen Aggregate belongs to the screen where it is defined and is not directly reusable as a screen-level aggregate elsewhere.
• Asynchronous and parallel execution: Multiple Screen Aggregates can run asynchronously and in parallel, helping the screen load data efficiently.

• A user connects with their Google account.
• Google associates the user’s SSH key with that identity.
• IAM roles determine whether they can log in.
• Access can be tied to a specific employee for auditing.
• Removing the IAM role removes their login access.

• compute.osLogin: regular OS login access
• compute.osAdminLogin: OS login with administrative privileges, typically including sudo

text 
enable-oslogin=TRUE

• Virus signatures: Primarily associated with antivirus or antimalware tools.
• Security agent deployment: Usually handled by endpoint-management or endpoint-security platforms.
• Vulnerability scanning: Performed by vulnerability-assessment tools, although a SIEM may ingest their results.

sql 
ALTER TABLE dbo.DimCustomer 
ADD CONSTRAINT PK_DimCustomer 
PRIMARY KEY NONCLUSTERED (CustomerKey) 
NOT ENFORCED;

1. ALTER TABLE dbo.DimCustomer
2. ADD CONSTRAINT PK_DimCustomer PRIMARY KEY NONCLUSTERED (CustomerKey)
3. NOT ENFORCED

• Confirm the required columns and data types
• Identify the target/label column
• Handle missing or duplicate records if present
• Transform data into a suitable format
• Split it into training and validation/test datasets

1. Prepare the training data
2. Select an algorithm or configure AutoML
3. Train the model
4. Evaluate its performance
5. Tune hyperparameters if needed
6. Deploy the model

• desktopwidth:i:1280
• desktopheight:i:1024

• Connection information: Controls connection-related settings, not the desktop resolution.
• Scaling plan: Controls when session hosts start, stop, or scale based on demand.
• Session behaviour: Controls behavior such as disconnect or logoff handling.
• Display settings: Specifically controls RDP display properties, making it the appropriate choice.

• A password cracker would normally be identified by behavior such as loading password hashes or attempting dictionary/brute-force attacks.
• The question describes a routine audit, not password recovery.
• Simply encountering a protected file during an audit is characteristic of a misconfigured or insufficiently privileged FIM.

• Existing measures
• Relationships
• Calculated columns
• Tables and fields
• Existing transformations in the model

• SharePoint folder: Reconnects to the source files. You would likely need to recreate the model and measures.
• Power BI dataflows: Primarily provide reusable data preparation and storage, but they do not automatically provide the existing semantic model and measures.
• Excel workbook: Connects directly to the source file and would also require rebuilding the model and measures.

• SharePoint search: Finds documents based on search terms, but it does not reliably identify every 32-character account number for DLP enforcement.
• A hunting query in Microsoft 365 Defender: Used for security investigations, not for defining sensitive content in a DLP policy.
• Azure Information Protection: Uses classification and labeling; it does not automatically identify this specific account-number pattern unless configured with a matching sensitive information type.
• RegEx pattern matching: Best suited for structured identifiers with a known format, such as account numbers, employee IDs, or policy numbers.

• Crowd-sourced evaluation: Requires recruiting and managing many evaluators.
• Human-worker evaluation: Provides human judgment but requires coordination, labeling instructions, and quality control.
• RLHF: Is a training and alignment process, not simply a low-overhead model comparison method.
• Automatic model evaluation: Quickly evaluates outputs using predefined toxicity or quality metrics with minimal manual effort.

• Due diligence: Investigating and evaluating a third party before entering into or continuing a business relationship.
• Attestation: An independent party provides assurance about an organization’s claims or controls. A SOC 2 examination is an example of an attestation engagement, but the analyst requesting and reviewing the report is performing due diligence.
• Internal audit: Reviews the organization’s own internal controls.
• Penetration testing: Actively tests systems for exploitable vulnerabilities.

• Accept: Acknowledging the risk without taking additional action.
• Transfer: Shifting the risk’s impact to a third party, such as through insurance or outsourcing.
• Mitigate: Reducing the likelihood or impact of the risk through security controls.
• Avoid: Eliminating the activity or process that creates the risk.

• IaaS: You deploy and manage virtual machines, including the operating system and installed software. Azure manages the physical datacenter, hosts, and networking infrastructure.
• PaaS: Azure manages the infrastructure and operating system, while you deploy and manage your application code. Examples include Azure App Service and Azure Functions.

1. Open the Microsoft Entra Connect wizard.
2. Select Configure.
3. Choose Customize synchronization options.
4. Authenticate when prompted.
5. On the Domain and OU filtering page, select:

1. Complete the wizard.

• Start-ADSyncSyncCycle only starts synchronization; it does not change which OUs are included.
• Manage federation configures federation settings, which are unrelated to OU filtering.
• Add-ADSyncConnectorAttributeInclusion concerns connector attributes, not selecting additional OUs for synchronization.

text 
interface GigabitEthernet0/0.10 
 encapsulation dot1Q 3

• .10 is the router’s local subinterface number.
• dot1Q 3 identifies the 802.1Q VLAN tag carried on the trunk.

text 
interface GigabitEthernet0/0.3 
 encapsulation dot1Q 3

text 
interface GigabitEthernet0/0.2 
 encapsulation dot1Q 2

• A — Too infrequent: Formal strategies are often reviewed only every three to five years. This can make the organisation slow to respond to rapid changes in technology, competitors, regulation or customer needs.

• C — Relies on predicting the future: The approach assumes senior managers can make reasonably reliable forecasts and assumptions. In uncertain environments, these assumptions may quickly become invalid.

• E — Can restrict radical innovation: Formal planning tends to favour existing information, established objectives and incremental improvements. This may discourage experimental or highly innovative strategies that emerge from lower levels of the organisation.

• B concerns the difficulty of achieving competitive advantage through core competences when competing with focused specialists. It is not specifically a drawback of the formal top-down planning process.
• D is an activity organisations should undertake—continuous monitoring—not a drawback of the approach itself.

• Supervised learning requires labeled examples to learn how to classify customers.
• Reinforcement learning learns through actions, rewards, and penalties, which is not the main task here.
• RLHF uses human feedback to improve model behavior, typically for generative AI, not customer segmentation.

• DB1 (500 TB): Azure Data Box Heavy

• Virtual machines: Azure Migrate

• Deterministic approve/reject results
• Strict adherence to predefined rules
• Consistent, reproducible outcomes
• Auditability and regulatory compliance

• Outer packet: 10.5.1.1 > 52.165.16.154
• IPv4 protocol field: 41, which indicates IPv6 encapsulation
• Inner packet: IPv6 traffic between 2001:... addresses

• MAC address spoofing: Changes Layer 2 addressing; it does not explain the IPv6-in-IPv4 structure.
• Steganography: Hides data inside files or media, not visibly in a tunneled protocol.
• Broken encryption: No evidence of encryption being cracked.
• On-path sniffing: Involves intercepting traffic, not sending data to an external host.

• User preferences change, so the model must learn from recent data.
• Keeping the existing data helps preserve knowledge of older patterns and provides a broader training distribution.
• Training only on new data can cause catastrophic forgetting and may overfit if the new dataset is small.
• The new data should not automatically be used as the test set, because it is needed for training and the model must still be evaluated on separate, representative data.

• Restrict a port to a defined number of MAC addresses
• Use statically configured or “sticky” MAC addresses
• Shut down or restrict the port when violations occur

• Load balancer: Distributes application traffic; it does not protect switch MAC tables.
• IPS: May detect suspicious traffic but is not the primary control for switch-port MAC limits.
• NGFW: Filters routed network traffic, not the switch’s CAM/MAC address table directly.

• A is correct: It establishes the revised dates and identifies the effect on the critical path.
• B is premature: Stakeholders should be informed after the impact has been assessed and the updated schedule is understood.
• C is not automatically appropriate: Fast-tracking may introduce additional risk and should be considered only after analyzing schedule options.
• D is premature: Adding resources may not resolve a dependency and could increase costs without addressing the actual constraint.

• Physical: Protects facilities or equipment, such as locks and security cameras.
• Technical: Uses technology, such as firewalls, encryption, or access controls.
• Operational: Covers day-to-day actions and processes, such as training, backups, or incident response.
• Managerial: Includes policies, standards, risk assessments, and procedures.

• A: Configure security groups — Customers define inbound and outbound rules for resources such as Amazon EC2 instances.
• B: Classify company assets — Customers decide how their data and assets should be classified and protected.

• C: Amazon S3 is a regional service; customers select a Region, not individual Availability Zones for buckets.
• D: AWS manages the infrastructure and patching for Amazon DynamoDB.
• E: AWS manages the infrastructure used by AWS Lambda; customers do not select EC2 instances for Lambda functions.