Cloud concept · Updated

Resource Efficiency - Resource efficiency is the practice of meeting workload requirements with the least necessary compute, storage, networking, and energy consumption.

In short

Resource efficiency means matching cloud capacity and architecture to useful work rather than provisioning for assumptions or leaving resources idle. It helps engineers improve performance per resource, control costs, and reduce environmental impact while preserving security and reliability requirements.

Cloud Concepts Hub →

6 key aspects of Resource Efficiency

Resource efficiency combines measurement, workload-aware resource selection, elastic capacity, and waste reduction within explicit security, reliability, and performance constraints.

01
Measure useful work per resource

Track resource consumption against outcomes such as successful requests, processed records, or completed orders. Combine utilization metrics with latency, errors, and throughput so that apparent savings do not hide degraded service.

02
Match resources to workload needs

Choose compute types, memory capacity, storage characteristics, and network capacity according to measured bottlenecks. Validate right-sizing recommendations with representative testing rather than relying on CPU averages alone.

03
Adapt capacity to demand

Use automatic scaling and appropriate schedules to reduce idle capacity while meeting demand. Account for provisioning delays, downstream limits, and the minimum capacity required for resilience.

04
Eliminate unnecessary work and retention

Reduce repeated computation, inefficient queries, unnecessary transfers, and data kept beyond its useful or required lifetime. Evaluate caching, compression, and lifecycle policies against their additional resource use and operational trade-offs.

05
Balance efficiency with service requirements

Preserve security controls, recovery capability, and availability objectives when reducing consumption. Spare capacity and redundant resources can be justified when they support explicit reliability requirements.

06
Separate consumption from price

Evaluate resource use, financial cost, and environmental impact as related but distinct measures. A pricing discount can lower expenditure without reducing consumption, while an architectural improvement may reduce both.

Cloud services for Resource Efficiency

Amazon CloudWatch provides the measurements needed to identify idle capacity and performance constraints, while AWS Compute Optimizer supplies recommendations for supported resources. Engineers can use those findings to select suitable Amazon EC2 instances and configure Amazon EC2 Auto Scaling to follow demand. AWS Lambda offers an alternative execution model for suitable event-driven work, and Amazon S3 lifecycle policies manage data according to access and retention needs. Together, these services support a measure, adjust, and validate cycle. Their use still requires explicit latency, availability, security, and recovery objectives; no individual service guarantees an efficient workload.

Amazon EC2Amazon EC2

Amazon EC2 provides instance families and sizes suited to different compute, memory, storage, and networking needs. Engineers improve efficiency by selecting and testing appropriate instances and stopping or terminating resources that are no longer needed.

Amazon EC2 Auto ScalingAmazon EC2 Auto Scaling

Amazon EC2 Auto Scaling adjusts the number of instances in an Auto Scaling group according to configured policies and schedules. Appropriate minimum capacity, scaling metrics, and instance warmup settings help reduce idle capacity without compromising responsiveness.

Amazon CloudWatchAmazon CloudWatch

Amazon CloudWatch supplies metrics, alarms, logs, and dashboards for observing workload demand and behavior. Memory utilization on EC2 generally requires an agent or custom metrics rather than relying only on default EC2 metrics.

AWS Compute OptimizerAWS Compute Optimizer

AWS Compute Optimizer analyzes utilization data and provides recommendations for supported resources, including EC2 instances. Recommendations help identify sizing opportunities but should be evaluated against workload constraints, representative demand, and recovery requirements.

AWS LambdaAWS Lambda

AWS Lambda runs functions in response to events without requiring customers to provision a continuously running server fleet. Function memory settings, execution duration, invocation patterns, and provisioned concurrency influence resource efficiency and cost.

Amazon S3Amazon S3

Amazon S3 supports lifecycle policies that transition eligible objects between storage classes or expire data. Matching storage to access and retention requirements can reduce unnecessary storage expenditure, while deleting unneeded data reduces retained data volume.

Resource Efficiency exam questions

Real practice exam questions on Resource Efficiency, with the correct answer and its explanation. Answer them with every option in the exam simulator.

MLA-C02 (beta)

A company runs a single-process ML feature-generation job on Amazon EC2 once each night. The job must finish within 4 hours, including instance startup and data loading. It requires 26 GiB of memory, cannot use swap, and cannot be distributed across instances. No compute resources are needed between runs. The workload is expected to remain unchanged for the next year.

Testing with the maximum expected dataset produced these results. Use the illustrative prices shown:

• m6i.xlarge: 16 GiB memory; $0.20 per On-Demand hour.
• r6i.xlarge: 32 GiB memory; completes the entire workflow in 3 hours; $0.30 per On-Demand hour.
• r6i.2xlarge: 64 GiB memory; completes the entire workflow in 2 hours; $0.60 per On-Demand hour.

A one-year Reserved Instance discount for an r6i.xlarge has an effective cost of $0.18 for every hour of the commitment, whether the instance runs or not. There are no other workloads that could use this discount. The job cannot tolerate Spot interruptions.

Which provisioning strategy meets the requirements at the lowest compute cost?

Show answerCORRECT ANSWER

Launch an On-Demand r6i.xlarge for each nightly job and terminate it when the job finishes.

The instance provides enough memory and completes the workflow within 4 hours. Each run costs $0.30 × 3 = $0.90. Terminating it afterward avoids idle compute charges, making this the lowest-cost configuration that meets all requirements.

AIP-C01

A company is replacing a large foundation model with a compact pretrained language model for an intermittent product-tagging workload. The compact model has passed quality and CPU-performance evaluations. Traffic drops to zero for several hours each day, and the company wants to run inference locally in AWS Lambda without maintaining an inference endpoint.

The required inference runtime and native dependencies occupy 8 GiB in an uncompressed container image. The model artifact occupies an additional 4 GiB. The runtime requires the complete model artifact on a local filesystem before loading it. Peak process memory during model loading and inference is 7 GiB. Tests confirm that downloading and initializing the model fit within Lambda initialization limits, and additional latency for new execution environments is acceptable.

Which deployment approach meets these requirements?

Show answerCORRECT ANSWER

Package only the runtime and dependencies in the container image. Store the model in Amazon S3, configure 8,192 MB of ephemeral storage and 10,240 MB of memory, and download the model to /tmp during initialization. Load the model outside the handler and reuse it for subsequent invocations in the same execution environment.

The runtime-only image fits within Lambda's image-size limit. Separately configured ephemeral storage accommodates the model artifact, and the memory allocation accommodates the measured peak process memory. Downloading and loading during initialization allows reuse within each execution environment without maintaining idle inference endpoints. New environments initialize independently, which the latency requirements permit.

DVA-C02

An application stores orders in an Amazon DynamoDB table with customerId as the partition key and orderTimestamp as the sort key. Each item is approximately 120 KB because it includes a detailed order document. A frequently used dashboard queries the 50 most recent orders for a customer but displays only customerId, orderTimestamp, status, and totalAmount. These attributes together occupy less than 1 KB per order. The application currently queries the table and retrieves all necessary pages. The dashboard can tolerate eventually consistent results. Which change will MOST reduce the read capacity consumed by dashboard requests while preserving the required results?

Show answerCORRECT ANSWER

Create a global secondary index with customerId as its partition key and orderTimestamp as its sort key. Project status and totalAmount into the index, and query the index for the 50 most recent orders.

The index contains the key attributes plus the projected summary attributes, so its entries are much smaller than the table items. Querying these smaller index entries consumes substantially less read capacity. The index supports the required customer and timestamp access pattern, and its eventual consistency meets the dashboard's requirements.

Practice Resource Efficiency in a full exam

Questions on Resource Efficiency in each practice exam, with every answer option and an explanation for each:

Start a free practice exam

What is Resource Efficiency

3 min read

Resource efficiency measures how effectively a workload turns infrastructure resources into useful outcomes. An efficient application might process more transactions per compute hour, store fewer redundant bytes, or transfer less data for each user request. The objective is not simply to use the smallest possible infrastructure footprint. It is to satisfy performance, availability, security, and business requirements without unnecessary consumption.

In the AWS Well-Architected Framework, resource efficiency is not a separate pillar. It is especially relevant to performance efficiency, cost optimization, and sustainability. Performance efficiency concerns using computing resources efficiently to meet requirements and maintaining that efficiency as demand and technology change. Cost optimization focuses on delivering business value at the lowest appropriate cost. Sustainability considers environmental impact, including reducing the resources required to deliver a workload. These objectives overlap, but a lower bill does not automatically mean lower resource consumption.

The other pillars establish important constraints and supporting practices. Operational excellence supports efficiency through monitoring, automation, experimentation, and continuous improvement. Reliability requires enough capacity and redundancy to withstand failures and recover within agreed objectives. Security requires protective controls even when those controls consume additional resources. Removing backups, encryption, logging, or redundant infrastructure solely to reduce consumption can undermine the workload rather than improve its design.

In practice, begin by measuring demand and defining useful outcomes. For a web application, track request rate, response latency, error rate, CPU utilization, memory pressure, and database activity. Compare these measurements with business metrics such as completed orders. Resource consumption per successful order is often more informative than average CPU utilization alone. Low CPU utilization might indicate oversized instances, but it might also reflect a memory-bound application or intentional recovery headroom.

Apply right-sizing and elasticity together. Right-sizing chooses suitable resource types and sizes; elasticity adjusts capacity as demand changes. For example, an online retailer can use smaller application instances during ordinary traffic and automatically add instances during a sale. Scaling policies must account for startup time, traffic spikes, downstream database limits, and minimum availability requirements. Maintaining some spare capacity can be a deliberate reliability decision rather than waste.

Architecture and software also affect consumption. Efficient queries, suitable indexes, batching, compression, and avoiding repeated computation can reduce required infrastructure. Caching can reduce database work but introduces memory consumption and consistency concerns. Storage lifecycle rules can move infrequently accessed objects to suitable storage classes or remove expired data, but retrieval charges, minimum storage durations, retention obligations, and recovery requirements must be considered.

Managed and serverless services can reduce idle provisioned capacity and infrastructure administration, especially for intermittent workloads. However, they do not automatically make an application efficient. Excessive invocations, inefficient code, unnecessary data transfers, and unsuitable execution patterns can still waste resources. Likewise, purchasing a discounted pricing commitment can reduce cost without changing the amount of infrastructure consumed.

Treat efficiency as an ongoing engineering process. Establish a baseline, change one significant factor, test under representative load, and compare results against service objectives. For exam questions, distinguish selecting and adapting suitable resources under performance efficiency, eliminating unnecessary spending under cost optimization, and reducing resource use and environmental impact under sustainability. A sound design balances these goals rather than maximizing utilization at the expense of every other requirement.

Learn Resource Efficiency in lessons

3 short lessons on Resource Efficiency in 3 certification courses.

Open the lessons