Cloud concept · Updated

Edge Computing - Edge computing places processing near data sources or users instead of relying exclusively on centralized cloud infrastructure.

In short

Edge computing distributes selected application tasks to devices, local gateways, nearby infrastructure, or provider-operated edge locations. It can reduce response latency, network traffic, and centralized processing costs, but introduces distributed infrastructure, security, and operational trade-offs.

Cloud Concepts Hub →

6 key aspects of Edge Computing

Edge computing balances proximity, data reduction, resilience, and centralized coordination against the cost and complexity of operating distributed resources.

01
Processing near sources and users

Place latency-sensitive or data-intensive work close to the devices or people that need it. Choose placement using measured network latency, connectivity reliability, and workload requirements rather than assuming every task belongs at the edge.

02
Selective data reduction

Filter, aggregate, compress, or analyze data before forwarding it to centralized systems. Retain sufficient raw data or representative samples for troubleshooting, auditing, and future analysis.

03
Complementary edge and cloud roles

Use edge resources for immediate decisions and regional infrastructure for shared services, durable storage, and large-scale analytics. Define ownership and synchronization rules for state that exists in both locations.

04
Explicit disconnected operation

Design local buffering, bounded storage, and recovery behavior when connectivity can fail. Test whether cloud-dependent credentials, configuration, or services prevent the intended offline operation.

05
Total-cost optimization

Compare reduced network and downstream processing demand against edge execution charges, hardware costs, idle capacity, and fleet operations. Edge processing is cost-effective only when the avoided costs and required benefits justify those expenses.

06
Distributed security and operations

Manage device identities, updates, monitoring, and physical exposure across many execution locations. Plan staged deployments and recovery from failed updates rather than treating remote nodes like a single centrally managed server.

Cloud services for Edge Computing

These services represent different edge placements rather than components that every deployment needs. AWS IoT Greengrass supports processing on devices and gateways, while AWS Outposts provides managed AWS infrastructure at customer sites. For web delivery, CloudFront Functions and Lambda@Edge execute selected request or response logic within the CloudFront delivery path. Regional services can remain responsible for durable storage, centralized analytics, and coordination. Choose the placement that meets latency, connectivity, and data-handling requirements, then compare execution charges, provisioned capacity, network costs, and operational effort against a centralized design.

AWS IoT GreengrassAWS IoT Greengrass

Runs application components on supported edge devices for local processing, messaging, and inference. Components can continue local work during connectivity interruptions when their dependencies and storage behavior are designed for it.

AWS OutpostsAWS Outposts

Extends supported AWS infrastructure and services to customer locations, enabling processing close to local systems and data. Its provisioned capacity and regional connectivity dependencies must be evaluated; it should not be assumed to provide fully disconnected operation.

Edge Computing exam questions

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

SAA-C03

A sports technology company provides real-time coaching feedback to athletes using devices connected to a telecommunications provider's 5G network. An EC2-compatible application processes each athlete's live motion data and returns personalized feedback. Results cannot be cached. The response-time requirement is 20 milliseconds, but testing shows that the network round trip to the nearest AWS Region alone takes 45 milliseconds. The provider supports AWS edge compute deployments in the athletes' metropolitan area.

The application also runs nightly model-training jobs that do not require low latency. Cost estimates show that these jobs cost less to run in the AWS Region than at the telecommunications edge. Training requires only small summaries from the live processing tier.

Which architecture meets the requirements MOST cost-effectively?

Show answerCORRECT ANSWER

Run live motion processing on EC2 instances in an AWS Wavelength Zone associated with the telecommunications provider. Send summaries to the parent AWS Region and run nightly model training there.

Wavelength places AWS compute within the telecommunications provider's infrastructure, avoiding the longer Regional network path for latency-sensitive processing. Keeping training in the Region uses the lower-cost capacity identified in the estimates and avoids provisioning unnecessary edge compute.

SAP-C03

A global retailer serves product pages through Amazon CloudFront. After catalog reorganizations, requests for retired product paths must receive HTTP 302 redirects to replacement paths, even when the retired pages are still present in CloudFront caches. Requests for all other paths must continue through the existing cache behavior unchanged.

The redirect mapping contains 10,000 entries totaling approximately 1 MiB, and each entry is less than 1 KiB. Catalog administrators update individual entries several times an hour. Updates must propagate without publishing new function code or invalidating cached pages. Normal propagation delay is acceptable. To minimize latency and avoid a Regional dependency, request processing must not make network calls to a Regional database or API.

Which solution meets these requirements?

Show answerCORRECT ANSWER

Associate a CloudFront Function that uses JavaScript runtime 2.0 with the viewer-request event. Associate a CloudFront KeyValueStore containing the redirect mapping with the function. Return a 302 response when the requested path has a mapping; otherwise, return the request. Update mapping entries through the KeyValueStore API.

The viewer-request event runs before cache lookup, so the function can redirect requests even when the retired page remains cached. CloudFront Functions can read an associated KeyValueStore at the edge without making a request-time call to a Regional database or API. Mapping updates propagate independently of function-code deployments, and the specified data fits the store's size limits.

Practice Edge Computing in a full exam

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

Start a free practice exam

What is Edge Computing

2 min read

Edge computing is an architectural approach that performs selected computations close to where data is generated or consumed. The edge might be an industrial gateway, a retail server, a connected device, or a cloud provider's distributed infrastructure near users. Its defining characteristic is processing location relative to the workload, not a particular service, hardware platform, or deployment model.

In cloud architectures, edge and centralized resources usually complement each other. Local components handle time-sensitive decisions, filtering, aggregation, or inference, while regional cloud services provide durable storage, fleet management, model training, and analysis across many locations. A factory gateway, for example, can evaluate vibration readings locally and upload summaries and abnormal readings rather than continuously transmitting every sample.

For cost-optimized compute design, identify what work can be eliminated or reduced before it reaches centralized systems. Filtering duplicate sensor events can reduce downstream ingestion, storage, and processing. Running inference near a camera can avoid continuously uploading video when only event metadata is needed. However, reducing device-to-cloud traffic does not automatically produce an AWS data-transfer saving: internet data transfer into AWS is generally free, while connectivity charges, downstream service charges, and data transfer out depend on the architecture.

Another application is lightweight request processing near web users. Edge functions can normalize URLs, issue redirects, or modify request headers before requests reach an application origin. Combined with appropriate caching, this can reduce origin requests and associated compute demand. Caching itself is not equivalent to executing application logic, although both techniques can be part of an edge architecture. Incorrect cache keys or personalization rules can expose user-specific data or reduce cache effectiveness.

The economic comparison must include the entire distributed system. Edge hardware, idle capacity, software deployment, monitoring, connectivity, physical maintenance, and replacement can outweigh savings from reduced centralized processing. Provider-managed edge functions have execution charges and runtime restrictions; local gateways require capacity planning and lifecycle management. A centralized, elastically scaled workload can be cheaper when data volumes are small, latency requirements are loose, or sharing compute across customers produces better utilization.

Intermittent connectivity is another reason to process locally, but offline operation must be designed explicitly. Applications need durable buffering, storage limits, retry policies, and duplicate-safe synchronization. Local processing does not guarantee continued operation if authentication, configuration, or required data still depends on an unavailable cloud endpoint. Similarly, proximity reduces network travel time but does not guarantee a hard real-time response.

Security and governance remain essential. Distributed devices need individual identities, least-privilege permissions, encryption, signed updates, and protection against physical compromise. Keeping raw data local can support privacy objectives, but logs, backups, and derived outputs must also satisfy residency and retention requirements. In architecture exam scenarios, favor edge processing when requirements emphasize local decisions, constrained connectivity, or reducing high-volume upstream data, then verify that its total cost and operational complexity are justified.

Learn Edge Computing in lessons

4 short lessons on Edge Computing in 2 certification courses.

Open the lessons