Design compute solutions
Specifying compute from workload requirements, and recommending VM, container, serverless and batch solutions.
- 15 min
- 5 concepts
- Try it yourself
- 4-question check
Learn
Choosing compute is about trade-offs: control versus management effort, steady versus bursty demand, and cost model. Microsoft's compute decision tree is the reference.
- 1
Specify components of a compute solution based on workload requirements
Work from requirements: control of the OS, existing packaging (code, containers, VMs), scaling pattern, start-up time, run duration, state, networking (VNet integration, private endpoints), hosting cost model and team skills. Then pick the service and size (CPU, memory, GPU, storage throughput).
Remember: Requirements first, service second.
- 2
Recommend a virtual machine-based solution
VMs suit lift-and-shift, custom software, or full OS control. Recommend the right family (D general, E memory, F compute, N GPU, B burstable), scale sets for elasticity, zones for availability, Azure Dedicated Host for compliance isolation, Spot for interruptible work, and reservations or savings plans for steady use. Azure Virtual Desktop for desktop delivery.
Remember: Steady 24/7 VMs: reserve them.
- 3
Recommend a container-based solution
Azure Container Apps: serverless containers and microservices with scale to zero, Dapr and KEDA, no Kubernetes management. AKS: full Kubernetes control for complex platforms. Azure Container Instances: simple single containers or burst capacity. App Service for containers: web apps packaged as containers. Store images in Azure Container Registry.
Remember: Microservices without managing Kubernetes: Container Apps.
- 4
Recommend a serverless-based solution
Azure Functions for event-driven code (Flex Consumption plan for fast scale and VNet integration, Premium for no cold start). Logic Apps for workflow and integration with hundreds of connectors. Durable Functions for stateful orchestrations. Serverless suits spiky, short-running tasks.
Remember: Code: Functions. Workflow with connectors: Logic Apps.
- 5
Recommend a compute solution for batch processing
Azure Batch runs large-scale parallel and HPC jobs on pools of VMs (including low-priority/Spot) that it schedules and scales. Alternatives: Container Apps jobs for containerised batch tasks, and Functions or Durable Functions for lighter fan-out work.
Remember: Thousands of parallel tasks on VMs: Azure Batch.
Compute at a glance
| Service | Control | Scale to zero | Typical use |
|---|---|---|---|
| Virtual Machines | Full OS | No | Lift and shift, custom software |
| AKS | Kubernetes | Nodes can scale down | Complex container platforms |
| Container Apps | Containers | Yes | Microservices, APIs, jobs |
| App Service | Code or container | No | Web apps and APIs |
| Functions | Code | Yes | Event-driven tasks |
| Batch | Job pools | Pools can shrink to zero | HPC and parallel jobs |
Try it
Walk Microsoft's compute decision tree.
This interactive part needs JavaScript.
In the portal
- Batch accountsAzure portal: Batch accounts > Create
- Function appsAzure portal: Function App > Create
- Container AppsAzure portal: Container Apps > Create
Key terms
Click a card to see the definition.
Check your understanding
4 questions. Score 80% or more and these skills are marked as confident in your AZ-305 study plan.
Which runs large parallel HPC jobs on managed VM pools?
Batch schedules work across pools.
Which serverless option is best for a workflow with many SaaS connectors?
Logic Apps has hundreds of connectors.
Microservices should scale to zero without managing Kubernetes. Which?
Container Apps hides Kubernetes.
Regulations require physical server isolation. Which?
Hardware dedicated to you.
Sources
- Microsoft Learn: Choose an Azure compute service
- Microsoft Learn: Azure Container Apps overview
- Microsoft Learn: Azure Batch overview
- Microsoft Learn: Azure Functions hosting options
Facts in this lesson are from Microsoft Learn. This site isn't affiliated with Microsoft.