Over-Provisioning — Azure Fundamentals (AZ-900)
You provisioned more capacity or redundancy than the scenario required. The exam rewards right-sizing.
Peak Headroom Isn't the Same as Right-Sized Compute
Requirement: handle unpredictable, bursty demand at minimum cost. Competing choices: reserved D-series VMs, Azure Spot VMs, serverless Functions. The deciding constraint is workload shape, not peak ceiling. Reserved instances optimize sustained, predictable load. Spot VMs serve fault-tolerant, interruptible jobs. Functions eliminate idle cost entirely for event-driven bursts. Sizing for peak without considering the pattern of demand is the trap — not a safe architectural choice.
The Scenario
A team needs storage for application logs. Logs are written continuously but only accessed during incident investigations — maybe once per quarter. You choose Premium Blob Storage for fast write performance. The correct answer is Standard Hot for recent logs (first 30 days) with a lifecycle management policy that moves data to Cool tier after 30 days and Archive after 90 days. Premium storage costs $0.15/GB/month; Standard Hot costs $0.018/GB/month; Cool costs $0.01/GB/month; Archive costs $0.002/GB/month. For 1TB of logs, Premium costs $150/month vs. a tiered approach averaging under $20/month.
How to Spot It
- •Azure Blob Storage tiers exist for different access patterns. Premium is for low-latency, high-transaction workloads (databases on disk). Hot is for frequently accessed data. Cool is for 30+ day retention. Archive is for 180+ day retention with hours of rehydration time. The exam tests whether you match the tier to the access frequency.
- •Azure Cosmos DB provisioned throughput at 400 RU/s (minimum) costs ~$23/month per container. If the scenario describes "occasional reads" or "low-traffic API," serverless Cosmos DB charges per RU consumed with no minimum, which can be pennies per month for light workloads.
- •Auto-scale and elastic tiers (Azure SQL Serverless, Cosmos DB autoscale, App Service auto-scaling) are the exam-preferred answer for unpredictable workloads. Fixed provisioned capacity is correct only when the scenario provides specific, stable throughput numbers.
Decision Rules
Whether a workload with predictable idle periods is better served by always-on provisioned compute billed per hour regardless of load, or by consumption-based compute that scales to zero and charges only for actual execution, where eliminating idle-hour spend is the dominant constraint.
Domain Coverage
Difficulty Breakdown
Related Patterns