Microsoft AZ-305: Compute and Application Architecture — Study Guide

By Laurynas Brazys · 11 August 2026 · #microsoft #az-305 #certification #exam-prep #study-guide

Part of the Microsoft Azure Solutions Architect Expert AZ-305 — Study Guide. Practice with verified answers in the Microsoft exam hub, or take timed practice tests on ExamRoll.io.

Overview

Azure compute spans infrastructure, platform, and serverless execution models. A solid architecture balances availability, cost, performance, operability, and security while matching the workload’s lifecycle and deployment topology. Understanding the capabilities and trade-offs of virtual machines, scale sets, App Service, Functions, container runtimes, orchestration platforms, batch scheduling, Service Fabric, and Azure VMware Solution enables precise, workload-appropriate designs.

Virtual Machines and Scale Sets

Azure virtual machines offer size families optimized for general purpose (Dv5, Dasv5), compute-optimized (Fsv2), memory-optimized (Ev5, Ebdsv5), storage-optimized (Lsv3 with local NVMe), high-memory (Mv2), HPC (HBv3/HC), and GPU (NCas, ND) workloads. Select sizes based on CPU/memory ratios, ephemeral storage needs, network throughput, and features like AMD vs Intel, accelerated networking, and ephemeral OS disks.

Availability sets distribute VMs across fault domains (rack/power) and update domains to reduce correlated failures within a single datacenter. Availability zones place VMs in physically separate datacenters within a region, offering higher SLA and true zonal redundancy at the cost of cross-zone latency. Proximity placement groups co-locate zonal or non-zonal resources to minimize latency within a tightly constrained physical topology—useful for chatty multi-tier systems (e.g., app tier to database).

Spot VMs use unused capacity at deep discounts with eviction on price or capacity pressure. Eviction policy (deallocate vs delete) and Max Price govern behavior. Spot is ideal for stateless, interruptible workloads such as CI builds, batch processing, and large-scale test fleets, not for critical stateful services.

Virtual Machine Scale Sets (VMSS) orchestrate identical VMs with unified lifecycle, health probes, automatic instance repairs, and zone spanning. Autoscaling uses Azure Monitor rules or schedules, reacting to metrics like CPU, memory (via custom metrics), queue depth, or request count. Predictive autoscale can pre-provision based on historical patterns. Orchestration modes include Uniform (homogeneous instances) and Flexible (heterogeneous, supports VM families and priority mixes). Rolling upgrades and automatic image upgrades use health checks and upgrade domains to drain, validate, and progress safely, with rollback on degraded health. VMSS integrates Spot capacity with capacity rebalancing and prioritization so scale-in/out favors eviction-prone instances first. Combine VMSS with zones for high availability, PPG for low latency, and application health extension for safe rollouts.

Web, Serverless, and Application Platform

Azure App Service provides managed hosting for Windows/Linux code and containers. Plans determine compute isolation and features:

Free/Shared for dev/test; Basic/Standard for production entry; Premium v3 for enhanced performance, VNet integration, hyper-thread isolation, and deployment slots; Isolated v2 (App Service Environment) for single-tenant, network-isolated hosting. Deployment slots enable blue-green and canary deployments with warm-up, slot-specific settings, traffic ramp-up, and zero-downtime swaps. VNet integration offers two dimensions: outbound regional integration via delegated subnets (route through NAT Gateway or custom routes) and inbound private access using Private Endpoints to disable public exposure while integrating with private DNS. App Service access restrictions, authentication/authorization, and managed identities reduce perimeter and credential risks. Scale up changes the plan tier/size; scale out increases instance count with autoscale rules tied to Azure Monitor.

Azure Functions delivers event-driven compute. Hosting options:

Consumption: per-execution billing with dynamic scale and cold-start trade-offs, max execution duration, and no VNet integration for Windows until Premium is used; ideal for bursty workloads.
Premium: pre-warmed instances eliminate cold starts, supports VNet integration, long-running executions, and predictable performance for enterprise integrations.
Dedicated (App Service Plan): runs on reserved App Service instances for steady workloads or co-hosting with web apps. Durable Functions introduce orchestrations, activities, and entities for long-running, reliable workflows implementing fan-out/fan-in, human interaction, and sagas with durable state and replay semantics. Bindings and triggers decouple code from services: HTTP, Timer, Service Bus, Event Hubs, Event Grid, Storage Queues/Blobs, Cosmos DB, and custom bindings. Concurrency, scaling behavior, and checkpointing differ by plan; design for idempotence and backoff policies.

Containers and Kubernetes

Azure Container Instances (ACI) provides fast, serverless containers without cluster management. A container group shares a network namespace, IP, and storage volumes; use sidecar patterns (e.g., logging agent next to an application container) and multi-container setups to co-locate tightly coupled processes. VNet integration via subnet delegation enables private address space, egress control, and private endpoints for dependencies; this facilitates burst or on-demand jobs from private environments. ACI is ideal for ephemeral tasks, CI/CD steps, and spiky workloads needing seconds-to-minutes startup times, not for complex service meshes or advanced scheduling.

Azure Kubernetes Service (AKS) manages Kubernetes control planes with optional Uptime SLA and offers system and user node pools. Use a minimal, tainted system node pool for core components and separate user pools per workload type (e.g., GPU, Spot, Windows). Integrate Spot node pools for cost savings on tolerant workloads with tolerations and PodDisruptionBudgets. The cluster autoscaler adjusts node counts in pools based on unschedulable pods; combine with the Horizontal Pod Autoscaler (HPA) and, for event-driven scaling, KEDA. Azure CNI assigns VNet IPs directly to pods, easing network policy, service discovery, and integration with on-prem via routable addresses but consuming more IPs; Azure CNI Overlay reduces VNet IP consumption while preserving CNI semantics. Kubenet uses NAT with fewer VNet IPs but adds SNAT complexity and limited advanced networking. Choose CNI for enterprise networking and policy, kubenet for smaller clusters with constrained address space, and Overlay for large-scale pod density without CIDR sprawl. Employ availability zones for node pools, surge upgrades for minimal disruption, and node/image upgrade channels to maintain posture.

Batch and Service Fabric

Azure Batch orchestrates large-scale parallel and HPC workloads without building your own scheduler. Pools define the compute: VM sizes (including GPU/HPC), images (Marketplace, custom, or Shared Image Gallery), and start tasks for environment bootstrapping. Application packages version your binaries; certificates and managed identities integrate secrets securely. Jobs group tasks, which can declare dependencies, use multi-instance coordination, and exploit task slots per VM. Job schedules automate recurring submissions. Auto-scaling formulas reference intrinsic metrics (e.g., $PendingTasks, $RunningTasks) and target counts (targetDedicatedNodes, targetLowPriorityNodes) with logic to react to queue depth, runtime averages, and cost constraints. Autopools create per-job pools for clean environments; persistent pools amortize initialization cost for frequent runs. Blend low-priority (Spot) nodes to reduce cost with failure-tolerant logic and checkpointing.

Azure Service Fabric is a distributed systems platform for microservices with strong state management. Stateless services scale horizontally and recover easily, while stateful services use Reliable Collections and replication to meet low-latency, in-memory state requirements with on-disk durability. Partitioning distributes load and state; replica sets maintain availability. Cluster reliability tiers (Bronze, Silver, Gold, Platinum) define seed node and quorum requirements, influencing failure tolerance and upgrade concurrency. Upgrade domains orchestrate rolling upgrades across the cluster by isolating failure impact; health policies and automatic rollback enforce safety. Service Fabric suits scenarios requiring high-throughput stateful processing, complex workflows, or when strong consistency and co-location with compute are paramount.

Azure VMware Solution and Compute Decision Framework

Azure VMware Solution (AVS) provides a fully managed vSphere, vSAN, and NSX-T stack on Azure bare metal. Private clouds run in Azure regions with vCenter access and familiar VMware tooling. Connectivity uses an AVS ExpressRoute circuit peered to Azure VNets via ExpressRoute Global Reach; from there, connect to on-premises via existing ExpressRoute or VPN. NSX-T networking provides micro-segmentation and north-south routing. HCX enables live vMotion, bulk migration, replication-assisted vMotion, and network extension for low-downtime datacenter evacuations and modernization phases. AVS is ideal for datacenter exit, app portfolios tightly coupled to VMware constructs, and transitional landing zones that later integrate with native Azure services over private links.

Choosing compute follows clear heuristics:

Use Azure VMs when you require OS-level control, custom images, specialized hardware (GPU/HPC), or legacy dependencies. Add availability sets/zones and PPGs to meet SLA and latency targets. Apply Spot for interruptible cost optimization.
Use App Service when hosting web APIs/sites with standard frameworks and you want managed platform, built-in DevOps slots, and autoscale without container/Kubernetes complexity. Choose Premium or Isolated when you need VNet integration and enterprise features.
Use Azure Functions for event-driven, bursty, or workflow-oriented tasks with minimal ops overhead; Premium eliminates cold starts and enables VNet/private access. Adopt Durable Functions for orchestrations and long-running activities.
Use ACI for ephemeral containers and jobs, rapid scale, and as a serverless execution target from pipelines or AKS virtual nodes. Choose VNet-integrated ACI for private workloads.
Use AKS for microservices, portability, advanced scheduling, service mesh, and standardized container DevOps at scale; select CNI model based on IP needs and networking policy requirements. Partition node pools by workload characteristics and availability zones, and enable autoscaling.
Use Azure Batch for parallel compute, rendering, Monte Carlo, and ETL workloads with scheduling, auto-scale economics, and low-priority capacity.
Use Service Fabric for high-throughput stateful microservices or when strong consistency, reliable actors, and precise control over upgrade/placement are required.
Use AVS to migrate VMware estates quickly, preserve operational models, and integrate with Azure networking and services as part of modernization.

Practical Problem Scenario

Contoso Retail is modernizing its e-commerce platform to handle flash sales with sub-10 ms latency between its API tier and a caching layer, meet a 99.99% SLA across a single region, minimize cost during normal demand, and migrate a legacy order management system still running on VMware. The solution must expose public APIs, process asynchronous events, and run nightly recommendation model scoring without impacting the live site.

Place core compute in a region with three availability zones and create a proximity placement group. Host the API tier in a zonal VMSS (Premium SSD v2, accelerated networking) inside the PPG for minimal latency to Redis Enterprise and a zonal SQL/Cache footprint. This choice delivers fine-grained performance control, PPG-level latency reduction, and zone-level resiliency for the hot path.
Configure VMSS autoscale based on request rate and custom queue depth metrics, with scheduled profiles for anticipated flash sales. Enable rolling upgrades with health probes and automatic image upgrades. This ensures elastic capacity and safe rollouts under extreme load while maintaining SLA.
Deploy the customer-facing web front end on Azure App Service Premium v3 with deployment slots for blue-green deployment. Integrate the app with a VNet for private egress to back-end APIs and data stores via Private Endpoints. App Service was chosen to offload platform management and streamline continuous delivery with safe swaps.
Implement asynchronous order processing with Azure Functions Premium using Service Bus triggers and Durable Functions for orchestrations (fan-out/fan-in for inventory checks, payment authorization, and notification). Premium eliminates cold starts, supports VNet integration, and Durable Functions provides reliable saga coordination.
Use Azure Batch for nightly recommendation model scoring across low-priority (Spot) nodes in a persistent pool preloaded with model dependencies. An autoscale formula targets node counts based on pending tasks and wall-clock predictions. Batch isolates the heavy compute workload from the operational store and optimizes cost with Spot capacity and checkpointed tasks.
Add ACI for bursty, ad-hoc data transformation jobs triggered by CI/CD and for tenant-specific report generation. Deploy container groups into a delegated subnet to keep all processing in private address space. ACI was selected for near-instant startup and zero cluster overhead for sporadic tasks.
Stand up Azure VMware Solution for the legacy order management system, migrating with HCX using replication-assisted vMotion and extending on-prem networks to AVS during the transition. Connect AVS to the application VNets via ExpressRoute Global Reach. AVS preserves operational tooling and accelerates datacenter exit without refactoring.
For future microservices, deploy AKS with zonal user node pools, a tainted system pool, and a Spot pool for tolerant workloads. Choose Azure CNI Overlay to conserve VNet IP space while retaining policy capabilities. Enable cluster autoscaler and HPA for layered scaling. AKS positions Contoso for container-native evolution where orchestration adds value.

This composition meets low-latency and high-availability demands via VMSS with PPG and zones for the hot path, simplifies web delivery with App Service, uses serverless and batch for asynchronous and batch compute efficiently, leverages ACI for ephemeral tasks, and employs AVS to migrate VMware-dependent systems—all integrated privately through VNets and Private Endpoints to maintain security and performance.

← Data Storage and Database Solutions · All domains · Networking and Connectivity →

Practice these questions → · Timed practice on ExamRoll.io →