# Right-Sizing

Right-sizing is matching cloud resource types, configurations, and capacity to workload requirements without unnecessary cost or unacceptable performance risk.

Cloud concept. Updated 2026-10-29. Page: https://blowstack.com/tools/cloud-concepts-hub/right-sizing

> Right-sizing uses observed demand and application requirements to choose appropriate compute, memory, storage, and throughput capacity. It reduces waste while preserving performance and reliability, and it requires regular review as workloads change.

Practice exams covering Right-Sizing: [ANS-C01 AWS Certified Advanced Networking - Specialty](https://blowstack.com/practice-exams/aws-certified-advanced-networking-specialty), [AIB-C01 (beta) AWS Certified AI Business Strategist](https://blowstack.com/practice-exams/aws-certified-ai-business-strategist), [CLF-C02 AWS Certified Cloud Practitioner](https://blowstack.com/practice-exams/aws-certified-cloud-practitioner), [SOA-C03 AWS Certified CloudOps Engineer - Associate](https://blowstack.com/practice-exams/aws-certified-cloudops-engineer-associate), [DEA-C01 AWS Certified Data Engineer - Associate](https://blowstack.com/practice-exams/aws-certified-data-engineer-associate), [DVA-C02 AWS Certified Developer - Associate](https://blowstack.com/practice-exams/aws-certified-developer-associate), [DOP-C02 AWS Certified DevOps Engineer - Professional](https://blowstack.com/practice-exams/aws-certified-devops-engineer-professional), [AIP-C01 AWS Certified Generative AI Developer - Professional](https://blowstack.com/practice-exams/aws-certified-generative-ai-developer-professional), [MLA-C02 (beta) AWS Certified Machine Learning Engineer - Associate](https://blowstack.com/practice-exams/aws-certified-machine-learning-engineer-associate), [SAA-C03 AWS Certified Solutions Architect - Associate](https://blowstack.com/practice-exams/aws-certified-solutions-architect-associate), [SAP-C03 AWS Certified Solutions Architect - Professional](https://blowstack.com/practice-exams/aws-certified-solutions-architect-professional).

## 6 key aspects of Right-Sizing

Effective right-sizing combines measured demand, appropriate resource selection, service-level safeguards, and repeated validation to reduce waste without impairing workloads.

1. **Evidence-based capacity selection**: Use utilization metrics and application performance indicators from representative operating periods. Include peak demand and failure conditions rather than relying only on averages.
2. **Multidimensional resource fit**: Evaluate CPU, memory, storage capacity, storage performance, and network limits independently. A workload may need a different resource family rather than a smaller version of its current configuration.
3. **Balanced cost and service objectives**: Remove unnecessary capacity while preserving latency, throughput, availability, and recovery requirements. Retain headroom when it has a measurable operational purpose.
4. **Continuous reassessment**: Review sizing as traffic patterns, software, and data volumes change. Combine periodic reviews with monitoring that detects sustained waste or resource pressure.
5. **Controlled implementation**: Validate recommendations through testing and staged deployment. Account for restart requirements, maintenance windows, and rollback before changing production resources.
6. **Distinct from scaling and discounts**: Automatic scaling adjusts capacity over time, while right-sizing establishes an appropriate resource configuration. Pricing commitments can complement right-sizing but do not correct an unsuitable configuration.

## Cloud services for Right-Sizing

CloudWatch supplies utilization and application signals, while Compute Optimizer turns supported resource telemetry into sizing recommendations. Engineers validate those recommendations and adjust EC2 instance configurations or RDS database capacity, accounting for compatibility, maintenance windows, and service objectives. EC2 Auto Scaling then helps an appropriately configured fleet respond to changing demand. These services support a repeated process of measuring, recommending, testing, changing, and observing; neither a recommendation nor an automatic scaling policy guarantees that a workload is correctly sized.

- **[AWS Compute Optimizer](https://blowstack.com/tools/cloud-services-hub/aws-compute-optimizer)**: Analyzes utilization and configuration data to recommend changes for supported resources, including Amazon EC2 instances. Treat its recommendations as candidates to validate against application requirements, peak demand, and operational constraints.
- **[Amazon CloudWatch](https://blowstack.com/tools/cloud-services-hub/amazon-cloudwatch)**: Provides metrics, dashboards, and alarms for evaluating utilization and validating sizing changes. Guest operating system memory utilization on EC2 generally requires the CloudWatch agent or another custom metric collection mechanism rather than default EC2 metrics.
- **[Amazon EC2](https://blowstack.com/tools/cloud-services-hub/amazon-ec2)**: Offers instance families and sizes with different compute, memory, storage, and networking characteristics. Right-sizing can involve changing instance size or family, subject to compatibility checks and any required stop/start or replacement procedure.
- **[Amazon RDS](https://blowstack.com/tools/cloud-services-hub/amazon-rds)**: Provides database instance classes and storage settings that can be adjusted to workload requirements. Evaluate database load, memory, connections, and storage performance together, and account for availability effects during modifications.
- **[Amazon EC2 Auto Scaling](https://blowstack.com/tools/cloud-services-hub/amazon-ec2-auto-scaling)**: Adjusts the number of EC2 instances in a group according to configured policies and demand. It complements right-sizing by varying fleet capacity, but the instance configurations and minimum capacity still need appropriate sizing.

## Right-Sizing exam questions

Real practice exam questions on Right-Sizing, with the correct answer and its explanation.

### Question 1 (AIB-C01 (beta))

**Question:**

A distribution company hosts an AI inventory assistant on an Amazon SageMaker AI real-time endpoint. The endpoint runs 12 identical instances continuously because warehouse employees submit a predictable surge of requests at the start of each shift. Each surge lasts 20 minutes. Representative load tests confirm that 12 instances are required during surges and two instances meet all performance and availability requirements between surges. New instances take approximately 10 minutes to become ready, and requests must receive immediate responses without queuing. The instance type and model must remain unchanged. Which plan should the business strategist recommend to reduce unused capacity while meeting these requirements?

**Correct answer:** Schedule capacity increases early enough for 12 instances to be ready before each shift surge, return to two instances afterward, and regularly review demand and performance.

**Explanation:** This plan matches instance count to demonstrated demand while accounting for startup time. It preserves peak performance and the tested availability baseline, removes unnecessary capacity between surges, and includes ongoing review as demand changes.

### Question 2 (MLA-C02 (beta))

**Question:**

A company plans to deploy a PyTorch model on an Amazon SageMaker AI real-time endpoint. The ML engineer has identified several compatible CPU and GPU inference instance types but does not know which provides the best price-performance.

The company has representative inference payloads and a required load-test profile consisting of a gradual traffic increase followed by a sustained peak. The selected configuration must maintain p95 model latency below 100 milliseconds at the required peak request rate. The engineer must automate benchmarking of the candidate instance types with this traffic profile and compare the costs of configurations that satisfy the performance requirements. The company does not want to build or maintain a custom load-generation framework.

Which approach best meets these requirements?

**Correct answer:** Run an advanced SageMaker Inference Recommender job with the candidate instance types, representative payloads, custom traffic phases, and performance thresholds. Compare costs among configurations that meet the peak throughput and p95 latency requirements.

**Explanation:** Advanced Inference Recommender jobs support customized load tests, including candidate endpoint configurations, traffic patterns, and performance thresholds. This enables workload-specific instance comparisons without maintaining a custom load generator. Cost should be compared only among configurations that meet the required throughput and latency.

### Question 3 (SAP-C03)

**Question:**

A company operates a fleet of Amazon EC2 workers that process large scientific datasets. Each worker requires 6 TiB of scratch storage on a single Amazon EBS data volume. The workload performs large, sequential reads and writes continuously for 18 hours each day. Testing establishes a requirement for sustained throughput of 300 MiB/s, including performance headroom. The solution must not depend on accumulated burst credits. HDD latency is acceptable, and the data can be regenerated if necessary.

Each EC2 instance supports sustained EBS throughput of 1,000 MiB/s. The application generates enough concurrent I/O to achieve the required throughput. On SSD volumes, the workload requires no more than 3,000 EBS IOPS.

In the company's AWS Region, st1 storage costs $0.045 per GiB-month. gp3 storage costs $0.08 per GiB-month, including 3,000 IOPS and 125 MiB/s of throughput. Additional gp3 throughput costs $0.04 per MiB/s-month. The company uses On-Demand pricing without discounts.

Which proposed configuration meets the requirements at the LOWEST monthly EBS cost per worker?

**Correct answer:** Use one 8 TiB st1 volume.

**Explanation:** st1 provides baseline throughput of 40 MiB/s per TiB, so an 8 TiB volume provides 320 MiB/s without relying on burst credits. It meets the capacity and sustained-throughput requirements for $368.64 per month (8 × 1,024 × $0.045), less than the qualifying gp3 configuration. Paying for extra HDD capacity is cheaper here than using SSD storage with additional provisioned throughput.

### Question 4 (CLF-C02)

**Question:**

A company runs millions of short-lived AWS Lambda invocations each day. As part of an environmental initiative, the company wants to reduce unnecessary memory allocation per completed invocation while preserving required response times. The team needs automatically generated memory-sizing recommendations based on historical utilization, rather than a questionnaire-based architecture review. Which service and AWS Well-Architected Framework pillar most directly match this requirement?

**Correct answer:** Use AWS Compute Optimizer recommendations to support the Sustainability pillar.

**Explanation:** AWS Compute Optimizer analyzes historical utilization and can recommend Lambda memory configurations. Using these recommendations to reduce unnecessary resource allocation per completed invocation most directly supports Sustainability when the stated goal is reducing environmental impact.

### Question 5 (SOA-C03)

**Question:**

A company wants AWS-generated rightsizing recommendations for 200 standalone Amazon EC2 instances in one account and AWS Region. Complete CPU, memory, disk, and network utilization metrics are available for a representative workload cycle. The application's vendor supports only the M6i, M7i, R6i, and R7i instance families. Other families must not appear as replacement candidates, even if they offer greater savings.

A CloudOps engineer must configure recommendation generation to enforce this restriction across the account, including for future instances, with the LEAST ongoing operational effort.

Which solution meets these requirements?

**Correct answer:** Opt in to AWS Compute Optimizer and configure account-level rightsizing recommendation preferences with preferred EC2 instance types restricted to the four approved families.

**Explanation:** Compute Optimizer supports preferred EC2 instance types in its rightsizing recommendation preferences. Restricting these preferences to the approved families constrains the replacement candidates. Applying the preferences at the account level provides a common configuration for existing and future instances without maintaining per-instance settings.

### Question 6 (AIP-C01)

**Question:**

A company runs a customer-service application on an Amazon Bedrock foundation model that supports Provisioned Throughput. The company must select a fixed number of model units for the next commitment period at the lowest cost while meeting its capacity requirements.

Monitoring shows peak demand of 300,000 input tokens per minute and 90,000 output tokens per minute. The capacity plan requires provisioned capacity for each token direction to be at least 120% of its observed peak demand. Load tests with representative requests confirm that each model unit simultaneously sustains 120,000 input tokens per minute and 24,000 output tokens per minute while meeting the application's latency objective. Throughput scales linearly across the configurations being considered, and each model unit has the same price.

How many model units should the company provision?

**Correct answer:** Provision 5 model units.

**Explanation:** The required capacities are 360,000 input tokens and 108,000 output tokens per minute. Input demand requires 360,000 / 120,000 = 3 units. Output demand requires 108,000 / 24,000 = 4.5 units, rounded up to 5. Five units satisfy both requirements and are the lowest-cost sufficient configuration.

### Question 7 (DEA-C01)

**Question:**

A company runs a memory-intensive AWS Glue Spark ETL job with a fixed worker count. The job must finish within 30 minutes. A data engineer benchmarks four configurations against representative peak-volume inputs and verifies that all configurations produce identical results. The measured durations include all billable runtime and remain consistent across repeated tests.

Configuration | Worker count | Worker type | Duration
A | 20 | G.1X | 28 minutes
B | 10 | G.1X | 42 minutes
C | 10 | G.2X | 23 minutes
D | 20 | G.2X | 17 minutes

A G.1X worker provides 1 DPU, and a G.2X worker provides 2 DPUs. Both worker types have the same price per DPU-hour. Which configuration should the engineer select to achieve the lowest compute cost per run while meeting the completion requirement?

**Correct answer:** Configuration C: 10 G.2X workers

**Explanation:** This configuration finishes within 30 minutes and consumes 10 × 2 × 23 = 460 DPU-minutes. This is the lowest consumption among the configurations that meet the deadline. The benchmark demonstrates that selecting larger workers with a lower worker count can reduce total processing cost.

### Question 8 (DOP-C02)

**Question:**

A company runs a data-transformation service on an Amazon EC2 Auto Scaling group. Each instance processes one job at a time by using a dedicated Amazon EBS gp3 volume. A job cannot be divided among instances. The business requires each job to finish within 5 minutes, including during sustained peak demand.

Short acceptance tests meet the requirement, but jobs take more than 10 minutes during hour-long demand peaks. Monitoring shows that CPU utilization remains below 30%, memory utilization remains below 40%, and network utilization is low. EBS throughput initially reaches 500 MiB/s but later remains near 125 MiB/s. Each volume is provisioned for 500 MiB/s, and its IOPS limit is not reached. The current instance type supports 125 MiB/s of baseline EBS throughput and up to 500 MiB/s through bursting.

Production-like tests establish that each worker needs at least 350 MiB/s of sustained EBS throughput to meet the deadline. The Auto Scaling group launches additional workers successfully, but individual jobs still miss the deadline. The company must resolve the issue without changing the application or the job format.

Which change should a DevOps engineer implement?

**Correct answer:** Select an instance type with at least 500 MiB/s of baseline EBS throughput and adequate CPU and memory. Validate sustained-load performance, update the launch template, and replace existing workers through an instance refresh that allows active jobs to finish.

**Explanation:** The instance's sustained EBS bandwidth is the bottleneck. An instance with at least 500 MiB/s of baseline EBS throughput can use the volume's provisioned throughput continuously and satisfy the measured 350 MiB/s requirement. Sustained-load validation and a controlled replacement address both sizing and rollout safety.

### Question 9 (DVA-C02)

**Question:**

A developer is evaluating smaller Amazon EC2 instance types for an API. During production peaks, clients submit 600 requests per second independently of earlier responses. The API must sustain this rate for 20 minutes with p99 response latency below 400 milliseconds and fewer than 0.1% failed requests.

In a production-like test environment, a script uses 120 simulated clients. Each client waits for its current request to finish before sending another request. On a smaller instance, the test reports acceptable latency and errors, but the request rate falls from 600 to 350 requests per second as responses slow down.

Which change should the developer make to obtain valid evidence that a candidate instance type meets the peak-demand requirements?

**Correct answer:** Configure the test to maintain an arrival rate of 600 requests per second with a representative request mix. Verify that the generators can sustain that rate, report requests they cannot issue, and measure client-side latency and failures for the full 20 minutes.

**Explanation:** The production arrival rate does not decrease when responses slow down. Maintaining the required arrival rate prevents the test from unintentionally reducing demand on an undersized instance. Generator-capacity checks and client-side measurements establish whether the candidate sustains the specified throughput, latency, and error requirements.

### Question 10 (SAA-C03)

**Question:**

A financial research company transforms a daily market-data export by using a proprietary Linux executable. Each run processes one export in a single process and builds a large in-memory lookup structure. The executable cannot partition a run across multiple hosts and cannot be modified.

The application currently runs on a general-purpose instance with 16 vCPUs and 64 GiB of memory. During transformation, CPU utilization averages 15%, memory is nearly exhausted, and the operating system continuously swaps pages to disk. Profiling shows that the application's peak working set, including operating system overhead, is approximately 180 GiB. Input reads and output writes are not bottlenecks.

Which change should a solutions architect recommend to address the transformation bottleneck?

**Correct answer:** Move the application to a memory-optimized Amazon EC2 instance with 16 vCPUs and 256 GiB of memory.

**Explanation:** The bottleneck is insufficient memory, as indicated by sustained swapping and low CPU utilization. A 256 GiB instance can accommodate the approximately 180 GiB working set with headroom, eliminating capacity-driven paging without requiring application changes.

## Practice Right-Sizing in a full exam

Questions on Right-Sizing in each practice exam, with every answer option and an explanation for each:

- [SAA-C03](https://blowstack.com/practice-exams/aws-certified-solutions-architect-associate): 33 questions
- [MLA-C02 (beta)](https://blowstack.com/practice-exams/aws-certified-machine-learning-engineer-associate): 17 questions
- [DVA-C02](https://blowstack.com/practice-exams/aws-certified-developer-associate): 9 questions
- [DEA-C01](https://blowstack.com/practice-exams/aws-certified-data-engineer-associate): 6 questions
- [SAP-C03](https://blowstack.com/practice-exams/aws-certified-solutions-architect-professional): 4 questions
- [CLF-C02](https://blowstack.com/practice-exams/aws-certified-cloud-practitioner): 4 questions
- [AIB-C01 (beta)](https://blowstack.com/practice-exams/aws-certified-ai-business-strategist): 3 questions
- [SOA-C03](https://blowstack.com/practice-exams/aws-certified-cloudops-engineer-associate): 3 questions
- [AIP-C01](https://blowstack.com/practice-exams/aws-certified-generative-ai-developer-professional): 3 questions
- [DOP-C02](https://blowstack.com/practice-exams/aws-certified-devops-engineer-professional): 1 question
- [ANS-C01](https://blowstack.com/practice-exams/aws-certified-advanced-networking-specialty): 1 question

## What is Right-Sizing

Right-sizing means selecting resources that fit a workload rather than simply choosing the smallest or cheapest option. An appropriately sized resource provides enough processing power, memory, storage performance, and network capacity to meet business requirements, with justified headroom for demand variation and failures. Both overprovisioning and underprovisioning are sizing problems: one wastes money, while the other can cause slow responses, failed jobs, or outages.

In the AWS Well-Architected Framework, right-sizing is most directly associated with Cost Optimization when the goal is eliminating unnecessary resource spending. It also supports Performance Efficiency by matching resource types and configurations to workload needs. The distinction depends on the outcome emphasized in a question: reducing the cost of an oversized instance points to Cost Optimization, while selecting the most suitable compute or storage technology points to Performance Efficiency. Reducing unnecessary resource consumption also supports Sustainability. Reliability concerns maintaining service through disruption and meeting availability requirements; Operational Excellence concerns operating, observing, and improving workloads; Security concerns protecting systems, data, and assets. Right-sizing must respect these requirements rather than treating cost reduction as the only objective.

Start with measurements collected over a representative period. Examine CPU utilization, memory pressure, storage latency, input/output operations per second, network throughput, request latency, and application error rates as relevant. Include business peaks, batch processing, seasonal demand, and failure scenarios. A low average CPU value alone does not establish that an instance is oversized: the application might require substantial memory, experience short CPU spikes, or be constrained by storage or network limits.

For example, a web application running on a large virtual machine might use little CPU and memory even during its busiest period. An engineer can test a smaller configuration, compare high-percentile response times and error rates, and deploy the change with a rollback plan. By contrast, an analytics worker with low CPU usage but high memory consumption may benefit from a memory-optimized resource rather than a uniformly smaller one. A database might need fewer virtual CPUs but still require its existing storage throughput.

Right-sizing is related to elasticity, but they are not identical. Right-sizing determines the appropriate resource shape and capacity for the work; elasticity adjusts capacity as demand changes. An automatically scaling fleet can still use oversized instances. Similarly, purchasing a discount commitment changes the price paid for capacity, not whether that capacity fits the workload. Validate sizing before making long-term purchasing commitments, and account for existing commitments when estimating actual bill reductions.

Changes involve trade-offs. Downsizing can reduce headroom, alter throughput limits, or require a restart or maintenance window. Reducing replica counts may weaken availability even if average utilization remains low. Use load tests, staged changes, monitoring, and rollback criteria to protect service objectives. Review sizing after application releases, traffic changes, and architectural changes; it is an ongoing engineering practice, not a one-time migration task.

## Learn Right-Sizing in lessons

17 short lessons on Right-Sizing in 9 certification courses.

- AIB-C01 (beta): Right-Sizing: matching AI resource capacity to demand (Premium), https://app.blowstack.com/lessons/right-sizing-matching-ai-resource-capacity-to-demand-aib-c01
- MLA-C02 (beta): Right-Sizing: matching capacity to workload requirements (Premium), https://app.blowstack.com/lessons/right-sizing-matching-capacity-to-workload-requirements-mla-c02
- SAP-C03: Right-Sizing: matching block storage capacity to demand (Premium), https://app.blowstack.com/lessons/right-sizing-matching-block-storage-capacity-to-demand-sap-c03
- CLF-C02: Right-Sizing: matching capacity to workload needs (Premium), https://app.blowstack.com/lessons/right-sizing-matching-capacity-to-workload-needs-clf-c02
- SOA-C03: Right-Sizing: matching compute capacity to workload demand (Premium), https://app.blowstack.com/lessons/right-sizing-matching-compute-capacity-to-workload-demand-soa-c03
- AIP-C01: Right-Sizing: foundation model inference capacity (Premium), https://app.blowstack.com/lessons/right-sizing-foundation-model-inference-capacity-aip-c01
