Cloud Computing Fundamentals
Where the services sit on the stack
Akamai Cloud Computing is an infrastructure-as-a-service platform: you rent virtual machines, storage volumes, and networking primitives running in Akamai data centers, and you manage the guest operating system yourself. Above that layer sit managed offerings such as LKE for Kubernetes and App Platform for application deployment, where Akamai takes on more of the operational burden. Expect questions about the shared responsibility split: Akamai owns the physical hosts, network fabric, and hypervisor; you own the guest OS, installed packages, application code, and identity configuration.
Core vocabulary
A region is a geographic location containing one or more data centers. A plan fixes vCPU count, memory, local storage, and a transfer allowance. Shared CPU plans let an instance burst but share physical cores with neighbors, while dedicated CPU plans reserve cores for that instance. Images are bootable disk templates, and StackScripts are scripts that run at first boot to configure a new instance. Elasticity means resizing plans or adding instances quickly; resilience means spreading workloads across data centers so a single failure does not take everything down.
Traps to watch
- Distributed regions expose a narrower catalogue of plans and smaller capacity than core regions, so "deploy the large dedicated plan everywhere" is usually wrong.
- A powered-off instance is still allocated and still billed; only deleting it stops charges.
- Scaling up (a larger plan) is not scaling out (more instances). Read the design goal in the question before choosing.
- Outbound transfer is pooled and inbound traffic is not metered, so "buy a bigger plan to get more inbound bandwidth" is a distractor.
Rule of thumb
Match the plan family to the workload: shared CPU for spiky, low-average workloads such as dev/test, small sites, and CI runners; dedicated CPU when sustained load or latency predictability matters; high memory when the working set is data rather than compute; GPU when the workload is model training or inference.