Task 4.1: Design compute solutions

• AAzure Kubernetes Service with the cluster autoscaler

  Incorrect

  AKS can do this, but the team would run and upgrade a Kubernetes cluster, which it wants to avoid.

• BAzure Container Apps

  Correct

  Container Apps runs containers on a serverless platform with KEDA-based scaling, scale to zero and revision traffic splitting, without cluster management.

• CAzure Container Instances, one container group per service

  Incorrect

  Container Instances runs single container groups without built-in autoscaling on queue length or revisions.

• DAzure App Service Web App for Containers on a Premium plan

  Incorrect

  App Service doesn't scale on queue length the way KEDA scalers do, and its plan instances don't scale to zero.

Azure Container Apps is a serverless container platform built on Kubernetes, with KEDA scale rules for HTTP and event sources, scale to zero, Dapr integration and revisions that split traffic, all without access to or management of the Kubernetes APIs.

• AThe Consumption plan

  Incorrect

  The Consumption plan is now a legacy plan without virtual network integration; new serverless apps should use Flex Consumption.

• BThe Flex Consumption plan

  Correct

  Flex Consumption is the serverless plan for new apps: it scales to zero, bills on use and supports virtual network integration.

• CThe Premium plan

  Incorrect

  Premium supports virtual networks but keeps at least one instance warm and billed, so it doesn't scale to zero.

• DA Dedicated App Service plan

  Incorrect

  A Dedicated plan bills its instances all the time, even when idle.

The Flex Consumption plan is Microsoft's recommended serverless hosting for new function apps. It adds virtual network integration and optional always-ready instances to event-driven scaling that goes to zero. The older Consumption plan is legacy; its Linux hosting retires on September 30, 2028, and Windows apps on it aren't affected.

• AAzure Functions on the Flex Consumption plan, one execution per frame

  Incorrect

  Functions suits short event-driven code, not 40-minute CPU-heavy renders with job queues and node pools.

• BOne large memory-optimized VM running all frames in sequence

  Incorrect

  One VM can't finish 20,000 frames overnight and pays for memory the renders don't need.

• CAn Azure Batch pool that autoscales and uses Spot nodes

  Correct

  Batch schedules and retries tasks across a pool that scales with the queue, and Spot nodes lower cost for interruptible work.

• DReserved VM instances for a fixed fleet of 200 render VMs

  Incorrect

  A fixed fleet sits idle outside the nightly run.

Azure Batch runs large-scale parallel and HPC jobs: it manages a pool of VMs, scales it, schedules tasks and retries them. Spot VMs in Batch pools are much cheaper and fit work that tolerates interruption.

• ACompute optimized

  Incorrect

  Compute-optimized sizes have a high CPU-to-memory ratio, the opposite of this profile.

• BMemory optimized

  Correct

  Memory-optimized sizes offer a high memory-to-CPU ratio for relational database servers and in-memory analytics.

• CGPU

  Incorrect

  GPU sizes target graphics, rendering and AI training, not a memory heavy database.

• DStorage optimized

  Incorrect

  Storage-optimized sizes focus on high disk throughput and local storage, not memory.

Azure groups VM sizes by type. Memory-optimized sizes provide a high memory-to-CPU ratio and suit relational database servers, medium to large caches and in-memory analytics.

• AA three-year savings plan for compute, with pay-as-you-go Windows licensing

  Incorrect

  A savings plan suits changing workloads with smaller discounts, and paying for Windows again wastes the owned licenses.

• BSpot VMs for all forty machines, plus Azure Hybrid Benefit

  Incorrect

  Spot VMs can be evicted at any time, which doesn't fit servers that must run around the clock.

• CThree-year reservations for the VM size, plus Azure Hybrid Benefit

  Correct

  Reservations give the deepest discount for stable size and region, and Hybrid Benefit removes the Windows license charge.

• DPay-as-you-go VMs that are deallocated outside business hours

  Incorrect

  The VMs must run 24 hours a day, so deallocating them isn't an option.

Reservations lock in a discount for a specific size and region and give the deepest savings when usage is stable; savings plans trade some discount for flexibility. Azure Hybrid Benefit lets Windows Server licenses with Software Assurance cover the VMs' OS cost.

• AAn App Service Environment v3 with an internal load balancer

  Correct

  An App Service Environment is a single-tenant deployment of App Service injected into your virtual network, with private-only access and high scale.

• BPremium v3 plans in the multitenant service with private endpoints

  Incorrect

  Private endpoints make apps private, but the compute stays on the shared multitenant platform rather than single-tenant.

• CAzure Container Instances in the virtual network

  Incorrect

  Container Instances runs containers without App Service features or the scale this many apps need.

• DStandard App Service plans with access restrictions

  Incorrect

  Access restrictions filter callers, but the apps stay multitenant and publicly addressable.

App Service Environment v3 is a fully isolated, single-tenant App Service deployment in a subnet of your virtual network. With an internal load balancer it's reachable only from the network, and it supports high instance counts.

• AConfidential VMs on shared hosts

  Incorrect

  Confidential VMs still run on hosts shared with other customers.

• BAzure Dedicated Host

  Correct

  A dedicated host is a physical server reserved for one subscription, with control over maintenance events.

• CA proximity placement group for the VMs

  Incorrect

  Proximity placement keeps VMs physically close for latency; it doesn't reserve the servers.

• DReserved VM instances for the VM sizes

  Incorrect

  Reservations are a billing discount and don't change where the VMs run.

Azure Dedicated Host provides physical servers that host VMs for one subscription only, giving hardware isolation at the server level and control over maintenance events.

• AEncryption at host on standard VMs

  Incorrect

  It doesn't protect data while the app processes it in memory.

• BAzure Dedicated Host for the VMs

  Incorrect

  A dedicated host doesn't protect memory from the host itself.

• CAlways Encrypted on the database the app reads from

  Incorrect

  Always Encrypted protects columns from the database server, not the app's memory on the VM.

• DAzure confidential VMs

  Correct

  Confidential VMs create a hardware-enforced boundary between the app and the virtualization stack, with no code changes.

Azure confidential VMs use hardware-based trusted execution to isolate a VM's memory and state from the hypervisor and host, so lift-and-shift workloads get data-in-use protection without code changes.

• ASeparate functions chained through storage queues with a status table

  Incorrect

  Hand-built chaining and a status table re-create the state and fan-in logic that Durable Functions provides.

• BDurable Functions with an orchestrator using fan-out/fan-in

  Correct

  Durable Functions orchestrations keep state, run activities in parallel, wait for all of them and retry failed ones.

• COne function that processes all chunks in a loop

  Incorrect

  A single long execution loses the parallelism and runs into execution time limits.

• DAn Azure Batch pool with one task per chunk

  Incorrect

  A Batch pool is VM-based compute outside the serverless app, and it doesn't coordinate the split, wait and combine steps as functions.

Durable Functions, an extension of Azure Functions, lets you write stateful orchestrations in code. Fan-out/fan-in runs many activity functions in parallel and waits for all results, with built-in checkpoints and retries.