CCNA Cloud Architecture and Design Questions

PaaS with a managed relational database provides a platform that handles infrastructure management while supporting relational databases with ACID compliance and strong consistency. This allows the company to focus on the application while ensuring the database meets its transactional requirements, making it the most appropriate model.

Amazon EFS is a fully managed, elastic NFS file system that supports the POSIX interface and can be mounted concurrently by thousands of EC2 instances across multiple Availability Zones within a Region. It automatically scales storage capacity up and down as files are added or removed, making it ideal for read-heavy shared workloads that require cross-AZ access.

This directly satisfies the requirement to survive a full data center failure. A global load balancer continuously probes regional endpoints and redirects user traffic to a healthy region when one fails. Because the workload is stateless and horizontally scalable, identical clusters can run in parallel without data synchronization concerns, providing both high availability and geographic redundancy.

Horizontal scaling adds EC2 instances, and CPU-based auto-scaling triggers new capacity only when load rises, then removes it afterwards. This matches spiky, stateless traffic efficiently and keeps cost low, unlike vertical scaling or always-on over-provisioning.

Archive storage offers the lowest per-gigabyte cost of Azure's blob tiers, designed for data retained for long periods and rarely accessed. It meets the seven-year retention requirement at minimal expense, unlike hot or cool tiers.

Azure Arc projects on-premises servers and Kubernetes clusters into Azure Resource Manager, enabling consistent management, policy, and monitoring across both environments while keeping data on-premises. Burst workloads can run in Azure connected to the same identity plane through Microsoft Entra ID, satisfying portability and centralized identity.

Keeping both tiers in one virtual network means traffic stays on the provider's private backbone and never crosses the public internet. A network security group or firewall rule scoped to the database subnet that permits only the application subnet's CIDR enforces least-privilege access at the subnet boundary, satisfying both the private-path and restricted-access requirements.

Rapid elasticity provisions and releases resources automatically as demand changes, satisfying the stem's requirement for scaling without human intervention. Measured service and on-demand self-service cover metering and user-initiated provisioning, but only rapid elasticity delivers the automatic, dynamic capacity adjustment described.

Horizontal scaling adds more VM instances to the pool and spreads requests across them, matching the stem's stateless web tier. Vertical scaling would instead resize a single existing VM, which cannot distribute load across multiple hosts.

Warm standby with continuous replication keeps a scaled-down environment running and replicates data continuously, giving an RPO near zero and failover well within two hours. This satisfies both the 15-minute RPO and 2-hour RTO at lower cost than active-active.

Multi-tenancy is the defining trait: the provider pools compute, storage and networking across customers, isolating tenants logically rather than physically. This shared-resource model satisfies the stem's public cloud characteristic, contrasting with single-tenant private deployments where hardware is dedicated to one organisation.

RPO defines the maximum tolerable data loss measured in time, determining how frequently backups or replication must occur. It directly satisfies the stem's requirement to establish recovery objectives by quantifying the acceptable gap between the last recoverable data point and the failure moment.

Asynchronous messaging decouples services by allowing them to communicate via messages without waiting for immediate responses. If a service fails, messages can be queued and processed later, providing graceful failure handling. This pattern promotes loose coupling and resilience, making it the most appropriate for the scenario.

An elastic IP address is a static public IPv4 address designed for dynamic cloud computing. It can be associated with any virtual machine in the account and moved between instances if a failure occurs. This allows clients that use the hard-coded IP to continue reaching the application even after the underlying virtual machine is replaced, satisfying the requirement for a stable public endpoint.

Warm standby satisfies the 4-hour RTO through pre-provisioned, scaled-down infrastructure that can be promoted quickly, while continuous data replication keeps the 1-hour RPO. It costs less than hot standby's fully mirrored active environment, avoiding the expense of duplicate capacity the RTO does not demand.

An Application Load Balancer operates at layer 7, terminating TLS and routing requests by URL path, host header or content. Network load balancers work at layer 4, forwarding packets by IP and port only, so they cannot inspect content or terminate SSL.

Amazon SQS is a fully managed message queuing service that enables asynchronous communication and decoupling of microservices. It supports standard and FIFO queues, dead-letter queues for handling failed messages, and message retention configurable from 1 minute to 14 days. This directly matches the requirements for a managed queuing service with dead-letter queue support.

Active-active runs identical workloads concurrently in both regions, each serving traffic, so a full region outage leaves the surviving region already handling production load. This differs from active-passive, where the standby region only takes over after failover, incurring downtime.

Object storage addresses the stem's scalability constraint by using a flat namespace with HTTP-accessible objects, so capacity grows without provisioning volumes. Block storage requires pre-sized LUNs and a filesystem, capping practical scale. This makes S3 suitable for unstructured data such as media and logs.

Customer-managed keys in a key management service let the organization own the key material and define access through policies. Revoking a user's permission to the key immediately removes their ability to decrypt data, and the company can rotate or disable keys itself. This meets both the encryption-at-rest and exclusive-control requirements.

Cloud SQL offers a high availability configuration that creates a standby instance in a different zone within the same region. It provides automatic failover and synchronous replication to the standby. This meets the requirement for a relational database with automatic failover and multi-zone replication within a region.

This design satisfies every stated constraint: the public subnet exposes web servers to the internet, the private app subnet is reachable only from the web tier via security groups, and the isolated database subnet lacks a route to an internet gateway, so no public IPs are needed.

Spreading the stateless application across two Availability Zones behind a load balancer removes a single point of failure. If one zone fails, the load balancer routes traffic to the surviving zone, so the design primarily improves high availability rather than scalability, elasticity or durability.

Archive tiers are engineered for long-term retention of data that is rarely accessed, offering the lowest storage price per gigabyte in exchange for retrieval times measured in hours. The scenario permits several hours of retrieval latency and prioritizes cost, so an archive tier aligns exactly with both the access pattern and the compliance retention period.

Hybrid cloud keeps sensitive customer data on-premises while extending workloads to the public cloud, directly satisfying the regulatory constraint. Unlike public or private cloud alone, it connects on-premises infrastructure with cloud services, letting the company migrate non-sensitive workloads without breaching data-residency rules.

IaaS delivers virtualised compute, storage and networking while leaving the guest OS, patching and middleware entirely under the customer's control. That directly satisfies the stem's requirement to retain control over the operating system and middleware, unlike PaaS or SaaS, which abstract those layers away.

VPN creates a private encrypted tunnel over the internet.

Block storage presents raw volumes that can be formatted with a file system and used by the database. It supports low-latency random read/write operations and allows the engineer to provision IOPS based on performance needs. Attaching the volume directly to the virtual machine ensures dedicated access and consistent performance, which is exactly what a high-performance database requires.

Horizontal auto-scaling adds or removes stateless instances in response to CPU thresholds, matching capacity to flash-sale demand without manual intervention. Because the application holds no session state, new instances can serve traffic immediately, satisfying the sudden-spike constraint efficiently. Vertical scaling would require restarts and hit fixed instance-size ceilings.

IaaS delivers only virtualised compute, storage and networking, leaving the guest operating system, middleware and applications entirely under customer management. PaaS and SaaS abstract the OS away, so IaaS uniquely satisfies the requirement for maximum control over both operating system and applications.

Deploying across multiple availability zones within a single region ensures high availability while keeping data within the country's borders, satisfying data residency. Availability zones are isolated data centers within a region, providing fault tolerance without crossing legal boundaries, making this the most critical design consideration.

PaaS delivers managed runtimes and middleware, so customers deploy code and manage applications while the provider handles the operating system, patching and runtime. That matches the constraint of no OS or runtime management, unlike IaaS, which leaves those layers to the customer.

IaaS delivers virtual machines, storage and networking while the customer manages the guest operating system, patching and middleware. That leaves the greatest control over the OS layer, unlike PaaS or SaaS, which abstract it away.

Amazon S3 is a durable, highly available object storage service designed for storing and retrieving any amount of data. It provides HTTP/HTTPS access and is ideal for backups and media files. S3 offers 99.999999999% durability and can be accessed globally, meeting the requirements.

A standby replica in another Availability Zone receives synchronous replication and can be promoted automatically when the primary zone fails. This provides a low recovery point and low recovery time because the standby is already in sync. It directly addresses the requirement that the database tier survive the loss of a single Availability Zone without manual intervention.

Distributing the web and application tiers across multiple availability zones ensures that if one zone fails, the remaining zones can continue to serve requests. A load balancer health-checks instances and routes traffic only to healthy targets. This eliminates single points of failure at the compute layer and is a fundamental practice for zone-resilient architectures.

Synchronous replication commits each transaction on both primary and standby before acknowledgement, so a failover loses zero committed data — comfortably inside the one-minute RPO. The hot standby is already running and current, allowing promotion within the 15-minute RTO. Asynchronous replication could not guarantee that one-minute data-loss limit.

A load balancer distributes incoming traffic across instances in multiple availability zones and health-checks them, routing around a failed zone. This directly delivers the fault tolerance the scenario demands by removing any single instance or zone as a point of failure.

AWS CloudFront caches static content at edge locations worldwide, directly satisfying the global low-latency requirement. Requests terminate at the nearest point of presence rather than the origin region, cutting round-trip time. Unlike regional services such as S3 alone, CloudFront's distributed edge network is purpose-built for this caching scenario.

IaC with Terraform allows defining infrastructure in declarative configuration files that can be version-controlled and applied consistently across multiple providers. This automates provisioning and configuration, reduces manual errors, and ensures repeatability, directly addressing the requirement for automated, consistent multi-cloud deployments.

In PaaS, the provider manages the runtime, middleware, and operating system, but the customer still writes, deploys, and patches the application code itself. This retained responsibility satisfies the scenario's requirement to identify what the customer still manages.

Egress charges apply whenever data leaves a provider's network, so cross-cloud replication and failover traffic incur fees that single-cloud designs avoid. This directly addresses the resilience-versus-cost trade-off the architect must weigh, since continuous synchronisation between clouds multiplies billable egress volume.

Workload Identity federates Kubernetes service accounts with Google Cloud IAM, allowing pods to obtain short-lived access tokens through the metadata server without any static key files. Each Kubernetes service account maps to a Google service account, so IAM roles can be scoped per workload, satisfying the keyless authentication and least-privilege requirements for the microservices deployment.

IaaS delivers raw compute, storage and networking while the customer retains control of the guest OS, middleware and applications. This satisfies the requirement for full control, unlike PaaS or SaaS, which abstract the operating system layer away from the customer.

An external Application Load Balancer is a global Layer 7 service that terminates TLS at Google's edge, routes requests to the nearest healthy backend, and supports managed certificates. Backends can be private instances, so they are not directly exposed to the internet, matching all stated requirements.

Stateless, horizontally scalable designs let the platform add or replace instances independently, so a failed node does not lose session state and traffic redistributes automatically. This satisfies the fault tolerance and elasticity benefit, unlike the single-server monolith's fixed capacity and single point of failure.

Horizontal scaling adds more server instances to the pool behind the load balancer, distributing increased traffic across them. Vertical scaling instead increases the CPU, memory or other resources of an existing instance, which does not match adding instances.

Active-active runs both Availability Zone instances concurrently, each serving traffic, so the loss of one zone leaves the other handling requests without failover delay. This satisfies the stem's requirement that the application remain available through a complete Availability Zone failure.

Placing the load balancer in a public subnet exposes only it to the internet, while web servers and database sit in private subnets without inbound internet routing. This tiered separation limits lateral movement and satisfies least-exposure best practise.

Horizontal scaling requires any instance to handle any request, but locally stored session state ties a user to one instance. Relocating session state to a shared database or cache externalises it, so additional instances can serve subsequent requests without affinity, enabling true horizontal scale-out.

Public subnets route to an internet gateway, so web servers placed there remain reachable from the internet. Private subnets have no such route, keeping database servers unreachable directly; outbound access via NAT still permits patching and updates without exposing them.

Spot instances exploit unused cloud capacity at steep discounts, satisfying the workload's tolerance for interruptions and its short nightly runtime. Because the batch job can resume after eviction, the risk of sudden reclamation is acceptable, making spot pricing markedly cheaper than on-demand or reserved capacity for this intermittent, fault-tolerant scenario.

Spot instances draw from spare capacity at steep discounts and can be reclaimed with little notice, which suits a stateless nightly batch job that tolerates interruptions. Reserved or on-demand pricing would bill for idle hours, failing the cost-reduction constraint.

Spot instances use spare cloud capacity priced far below on-demand rates, and they can be reclaimed with little notice. The job tolerates interruptions and resumes from checkpoints, so eviction causes no data loss. This fault tolerance is exactly the constraint that makes spot pricing the cheapest viable option.

A warm standby keeps a scaled-down but functional copy of the environment running in the secondary Region, and continuous replication keeps data within the five-minute recovery point. Automated failover orchestration promotes the standby and redirects traffic with minimal human action, meeting the fifteen-minute recovery time. This balances cost and speed better than hotter or colder options.

Aurora Global Database is designed for global applications. It consists of a primary cluster in one region and up to five secondary clusters in other regions. Replication is storage-based and typically has less than one second of latency. Secondary clusters can serve read traffic with low latency, and failover to a secondary region can be performed quickly. This meets the requirements for read-heavy workloads, low latency, and high availability.

AWS Fargate is a serverless compute engine for containers that eliminates the need to manage servers. An Application Load Balancer provides advanced routing and integrates with ECS. Configuring the ECS service across multiple Availability Zones ensures high availability. This solution scales rapidly and reduces operational overhead, directly addressing the requirements.

Azure Files offers fully managed file shares in the cloud that are accessible via SMB and NFS protocols. It supports simultaneous mounting by multiple VMs, including across different availability zones when using zone-redundant storage. This makes it suitable for lift-and-shift applications that need shared file storage.

Enforcing TLS on the database listener with mandatory certificate validation ensures the channel between application and database is encrypted and that the application verifies the database identity, preventing interception or spoofing. This directly produces the cryptographic evidence auditors require for encryption in transit between the two tiers.

FaaS runs event-triggered code without provisioning or managing servers, matching the startup's requirement exactly. The provider handles all infrastructure; the developer deploys only functions. IaaS and PaaS still involve server or runtime management, so they fail the no-server-management constraint.

IaaS provides the raw compute, storage and networking primitives while leaving the guest operating system, middleware and installed software entirely under the customer's administration. That satisfies the stated requirement for full control over the OS, software and security configurations, which PaaS and SaaS abstract away.

Hybrid cloud keeps sensitive workloads on-premises while extending other workloads to public cloud, directly satisfying the regulatory constraint that sensitive data must remain on-premises. Neither pure public nor private cloud alone meets both the migration goal and the data-residency requirement.

Distributing EC2 instances across two or more Availability Zones ensures that an AZ-wide failure only removes part of the fleet, leaving capacity in surviving zones. This directly satisfies the stem's requirement to withstand the failure of an entire Availability Zone.

Placing only the web tier in a public subnet limits internet exposure to the presentation layer, while application and database tiers remain unreachable directly. Routing through NAT gateways or load balancers enforces tiered segmentation, satisfying least-privilege network access.

File storage exposes SMB or NFS shares, so several virtual machines can mount the same file system concurrently. Block storage attaches to a single instance, and object storage lacks the shared hierarchical file semantics required.

Amazon Aurora Global Database is designed for multi-region applications, providing a primary region that is writable and secondary regions that are read-only with typical replication lag under one second. It uses dedicated replication infrastructure for low latency and supports fast failover. This meets the requirements for a writable primary and low-lag readable secondary region.

CDN caches content at edge locations, reducing latency.

Spreading resources across multiple availability zones within one region gives fault isolation: a single zone failure leaves the application running elsewhere. This directly satisfies the stem's fault-tolerance and high-availability constraint while remaining inside a single region.

Auto-scaling groups add or remove identical stateless instances horizontally as CPU utilisation rises and falls, matching capacity to demand. This satisfies the cost-effectiveness constraint because instances are billed only while running, unlike vertical scaling, which requires permanently larger, pricier instances.

File storage exposes SMB or NFS shares, letting multiple VMs mount the same file system concurrently. Block storage attaches to a single instance, and object storage lacks a POSIX NFS interface. NFS concurrency is therefore satisfied only by file storage.

Amazon S3 Standard-IA is designed for data that is accessed less frequently but requires rapid access when needed. It offers the same low latency and high throughput as S3 Standard, with 11 nines of durability, and has a lower storage price. It is the most cost-effective choice for infrequently accessed data that must be immediately available, though it has a minimum storage duration charge of 30 days.

Spot instances exploit spare EC2 capacity at steep discounts, suiting the stateless, interruptible workload described. The Application Load Balancer distributes traffic across instances, so a reclaimed Spot instance causes no data loss, and variable demand is met elastically. This satisfies the stem's cost-effectiveness and tolerance-for-interruption constraints better than On-Demand or Reserved pricing.

Direct Connect and ExpressRoute provide dedicated private circuits between on-premises infrastructure and cloud providers, bypassing the public internet entirely. This satisfies the stem's requirements for high bandwidth, low latency, and privacy. Unlike site-to-site VPNs, which tunnel over the internet, these services deliver consistent performance through a direct physical link.

IaaS delivers virtualised compute, storage and networking while leaving the guest operating system, middleware and applications to the customer. This satisfies the requirement for maximum control, since the customer patches, hardens and configures the OS and software stack themselves.

A managed network file system designed for concurrent access provides the shared POSIX namespace the cluster needs, and its throughput often scales with provisioned capacity, which suits a multi-week, high-throughput training job. It supports many clients mounting the same file system simultaneously, so all virtual machines see consistent file metadata and can read and write in parallel.