Cloud Fundamentals

How Cloud Computing Works (With Diagram)

How cloud computing works, from browser to data center: virtualization, orchestration, storage, metering and regions, with a simple text diagram.

On this page
  1. The basic picture
  2. The front end: what you use
  3. The back end: what the provider runs
  4. Virtualization: the core enabler
  5. Containers and orchestration
  6. How a resource gets created
  7. Multi-tenancy and isolation
  8. Storage types: block, file and object
  9. Load balancing and scaling
  10. Regions and zones
  11. Metering and billing
  12. Serverless
  13. A short timeline
  14. Common misconceptions
  15. Security note
  16. Frequently asked questions

Cloud computing works by splitting a service into two sides. The front end is what you use: a browser, mobile app, command-line tool or software development kit (SDK). The back end is the provider’s data center, where physical servers, storage and network equipment run the software. The two sides communicate over the internet, usually with HTTPS requests and APIs. Inside the back end, virtualization and orchestration turn physical hardware into resources you can create and release on demand.

The basic picture

  YOUR DEVICE                  INTERNET                  PROVIDER DATA CENTER
  -----------                  --------                  --------------------
  Browser                                                Load balancer
  Mobile app   ----------->   HTTPS requests   ---------> Applications
  CLI or SDK                                              Virtual machines
                                                          Databases and storage
                                                          Orchestration and security
                                                          Physical servers and network

The device does the least work. It sends requests and displays results. The data center does the heavy lifting.

The front end: what you use

The front end is the part you interact with. It can be:

  • A web browser, such as the cloud provider’s management console
  • A mobile app or desktop client
  • A command-line interface (CLI) or software development kit (SDK)
  • A thin client that does little more than display what the back end sends

Your device needs an internet connection and the right software. It does not need to store or process your application’s data.

The back end: what the provider runs

The back end is built in layers:

  • Physical infrastructure: servers, storage arrays and network equipment in data centers
  • Virtualization: software that divides each physical server into several isolated virtual machines
  • Management and orchestration: software that provisions, monitors and releases resources
  • Services: databases, load balancers, message queues, AI tools and other building blocks
  • Security: identity, access control, encryption and network protection
  • Billing and metering: measurement of usage and the records that produce invoices

Each layer hides the one below it. You usually work with services, not with the servers themselves.

Virtualization: the core enabler

A hypervisor runs on a physical server and divides its CPU, memory and storage into several virtual machines (VMs). Each VM runs its own operating system and applications, and it is isolated from the others. Hypervisors used in cloud data centers include KVM, Xen, VMware ESXi and Microsoft Hyper-V. AWS’s Nitro system is built on KVM.

Virtualization in cloud computing is necessary, but it is not the whole cloud. Cloud adds self-service, metering and automation on top of it.

Containers and orchestration

Containers package an application with its dependencies. Unlike a VM, a container shares the host operating system’s kernel, so it is lighter and starts faster. Docker, released in 2013, popularized containers. Kubernetes, open-sourced by Google in 2014, schedules and manages containers across many machines.

Containers and VMs often work together. A company might run a Kubernetes cluster on a set of VMs, for example.

How a resource gets created

When you ask for a server, storage bucket or database, the request travels through a sequence of steps:

  1. You submit the request through the web console, CLI, SDK or API. Most APIs use REST over HTTPS.
  2. The provider’s orchestration software checks your permissions and quotas.
  3. The orchestration software chooses physical capacity with room for the request.
  4. A hypervisor creates the VM, or the platform allocates the container, storage or database.
  5. Networking and security rules are applied, and you receive an address or endpoint to connect to.

This usually takes seconds to minutes, and no provider staff member has to approve each request. That self-service model is what makes the cloud feel instant.

Multi-tenancy and isolation

Many customers share the same physical hardware. Each customer’s resources are isolated logically, through separate VMs, network rules, encryption and access controls. Isolation is enforced by software and configuration, not by separate hardware for each customer.

This sharing is what makes the cloud efficient, and it is why access controls and configuration matter so much. See the shared responsibility model for who secures what.

Storage types: block, file and object

Cloud storage comes in three main types:

  • Block storage acts like a disk attached to a VM. Examples include Amazon EBS.
  • File storage provides shared folders that several machines can mount. Examples include Amazon EFS and Azure Files.
  • Object storage stores files as objects with unique keys, accessed through an API. Examples include Amazon S3, Azure Blob Storage and Google Cloud Storage.

Amazon S3 is designed for 99.999999999% (11 nines) durability. That figure describes how unlikely it is that an object is lost. It is not a promise that the object is reachable at every moment, which is what availability describes.

Load balancing and scaling

A load balancer spreads incoming traffic across several instances, so no single machine is overloaded. Auto scaling adds instances when demand rises and removes them when it falls. Together, these let an application handle changing traffic without a person adjusting servers. For more on this, see what is elasticity in cloud computing?

Regions and zones

Providers organize their infrastructure into regions, which are geographic areas. Each region contains availability zones, which are isolated data centers with their own power and networking. Where you place a resource affects latency, data rules and cost. See cloud regions and availability zones explained.

Metering and billing

Cloud providers measure what you use and bill for it. Examples vary by service:

  • AWS bills Linux EC2 instances per second, with a 60-second minimum.
  • Storage is usually billed per gigabyte per month.
  • Data transfer out of the provider is usually billed per gigabyte.

Prices change and vary by region, so check current rates. Metering is also what makes the pay-as-you-go model possible.

Serverless

Serverless services, such as AWS Lambda (launched in 2014), run your code in response to events. You do not provision or manage the servers. AWS Lambda functions can run for up to 15 minutes per invocation.

Serverless still runs on servers. The difference is that the provider manages them, so you do not have to.

A short timeline

  • March 2006: Amazon S3 launches.
  • August 2006: Amazon EC2 launches.
  • 2008: Google App Engine launches.
  • February 2010: Microsoft Azure becomes generally available.

These dates show how quickly the model matured. Most of the layers described above are the result of that work.

Common misconceptions

  • The cloud is abstract. It is physical servers in data centers.
  • The cloud is the internet. The internet is the transport. The cloud is the remote computing you reach over it.
  • Virtualization is the cloud. Virtualization is one layer. The cloud adds self-service, metering and automation.
  • Data is stored everywhere. Data lives in the regions you choose, unless you configure it to move.
  • Serverless has no servers. The provider runs the servers for you.

Security note

Misconfigured access, such as a public storage bucket, is a common cause of cloud data exposure. Check access settings and use multi-factor authentication for every account with administrative rights.

Frequently asked questions

What happens when I create a cloud server?

Your request goes to the provider’s API, where it is checked against your permissions. Orchestration software chooses capacity in a data center, a hypervisor creates the virtual machine, and networking and security rules are applied. You then connect to the server’s address.

Is the cloud the same as the internet?

No. The internet is the network that carries your requests. The cloud is the computing and storage at the far end.

Do I need to understand the back end to use cloud computing?

Not to get started, but a basic picture helps. It explains why regions, storage types and billing behave the way they do. For the types of cloud a business can choose, see cloud deployment models.

Keep exploring

Guides from other topics

All topics →