openskills.info
Course catalog
AWS Compute logoCourse Preview

AWS Compute

AWS compute services provide on-demand processing capacity in Amazon's cloud. They range from virtual machines (EC2) and containers (ECS, EKS) to serverless functions (Lambda), each trading control for operational simplicity depending on the workload.

itCloud computing

Don't Panic — AWS Compute

AWS Compute is a menu of execution models, which is a polite way of saying AWS has several ways to run your code and each one brings a different pile of chores. You can rent a virtual server, schedule containers, run a function after an event, or submit a finite job to a queue. The useful question is not which service name sounds familiar. It is what you need to control and what you are prepared to operate.

Amazon EC2 is the control-rich choice. An AMI starts a virtual server, an instance type shapes its resources, and the rest of the arrangement is firmly your business: guest operating-system updates, network access, credentials, application software, capacity, and replacement. One instance is one failure domain, which is less a philosophical observation than a warning label. The Intro explains how launch templates, Auto Scaling groups, healthy targets, and load balancing turn replaceable instances into a service.

Containers split the question in two. ECS and EKS decide how container work is orchestrated; EC2 and Fargate decide where that work gets capacity. Fargate removes the underlying instance fleet, not the image, task permissions, network paths, storage choices, or application behavior. If that sounds like a distinction designed by a committee, it is still the distinction that keeps an incident from becoming an archaeology project. The Cheatsheet has the compact comparison when the names start to resemble a bowl of initialisms.

Lambda narrows the contract again. It runs an independent function invocation after an event or API call, so durable state belongs elsewhere and a persistent host process belongs somewhere else entirely. AWS Batch takes the other timing path: finite work enters a queue and is scheduled for throughput, retries, and resource use rather than interactive request latency. Lightsail is the smaller packaged-web route when the full EC2 architecture surface would be more choices than the project needs.

The surprise is that managed compute does not remove architecture. It moves responsibility upward and changes the shape of the constraints. Cost also follows the allocation model: EC2 capacity, Fargate task resources, Lambda requests and execution resources, Batch resources, and Lightsail bundles do not fail or bill in the same way. Read the Slides for the selection path, then use the Field Notes when the neat diagrams have met queues, overload, and an Availability Zone that has chosen an inconvenient afternoon.

When you need to put the picture to work, use the Reference tab for AWS documentation and the Landscape for the compute products behind the choices. Keep the mental model small: choose the smallest execution unit that fits the workload, then check identity, network access, state, failure behavior, observability, and cost. The cloud is still a computer. It has merely become very good at sending you a receipt for the parts you forgot to think about.

Where this skill leads

Relevant careers

See how this topic contributes to broader role-level skill maps.

Sources