Skip to content
Knowledge Hub/Article
Article

What Is a Cloud Server? How It Works and How to Choose One

A cloud server is a virtual server rented from a cloud provider and paid for as you go. How it differs from a VPS, pricing models and how to choose one.

Oğuzhan Gerçek··8 min read
What Is a Cloud Server? How It Works and How to Choose One

Short answer: A cloud server is a virtual server that runs in a cloud provider's data center and is rented over the internet. You choose the processor (vCPU), memory, disk and network capacity, and you pay for what you use. The provider runs the hardware; the operating system and the software on it stay with you. The right choice depends on the workload profile, the region where the data will live, the pricing model and who will operate the server.

What is a cloud server?

AWS defines every server launched in EC2 (an instance) as a virtual server in the AWS Cloud. According to Microsoft, Azure virtual machines give you the flexibility of virtualization without having to buy and maintain the physical hardware that runs them. Google Cloud, for its part, notes that a Compute Engine instance can be either a virtual machine or a bare metal instance.

A cloud server is the core building block of the IaaS model. In NIST's definition, the consumer in this model does not manage the underlying infrastructure but does control the operating system, storage and deployed applications. We cover the cloud's service models in our cloud computing article, servers in our server article and virtualization in our virtual machine article, and there is a short definition in our glossary. A cloud server can run in a public cloud or in a private cloud dedicated to a single organization.

How does a cloud server work?

According to the AWS EC2 documentation, these are the parts of a cloud server:

  • Image (AMI): A ready-made template that contains the operating system and additional software.
  • Instance type: A specific combination of CPU, memory, storage and networking capacity.
  • Disk: Block storage that keeps data persistently (Amazon EBS). Some types also have a temporary disk (instance store) whose contents are deleted when the instance is stopped.
  • Network and access: A virtual firewall (security group) that sets which ports and IP ranges can reach the instance, and a key pair for logging in.

An instance is launched in a region. According to AWS, each region has several isolated Availability Zones, and spreading instances across more than one zone protects the application from the failure of a single location.

The bill is itemized too. On AWS, a stopped instance incurs no compute charges, but EBS volumes and Elastic IP addresses are charged regardless of the instance's state. On Azure, a VM shut down from inside the operating system keeps its place on the hardware and is still billed; compute charges stop only once the VM is deallocated.

Instance families and vCPUs

Providers group their servers into families. In AWS naming, the first letter of c7gn.xlarge is the series (c: compute optimized), the number is the generation, the next letters are options (g: AWS Graviton processor, n: network and EBS optimized), and what follows the period is the size. General purpose series start with M, memory optimized series with R. Azure and Google Cloud likewise offer families optimized for general purpose, compute, memory, storage and accelerators such as GPUs.

A vCPU is not always a physical core. On AWS, instance types that support simultaneous multithreading (SMT) count each thread as a vCPU: an m5.xlarge presents its 2 cores as 4 vCPUs. In Google Cloud's N4A series, by contrast, one vCPU is equivalent to an entire physical core. Keep the difference in mind when sizing a workload moved from a physical server.

Cloud server vs physical server, VPS and shared hosting

As AWS explains it, hosting options form a spectrum:

  • Shared hosting: Sites share the same physical server and compete for resources; the load from neighboring sites can affect your performance.
  • VPS: A virtual server that uses a share of a physical server, with the resources in that share reserved for you.
  • Dedicated server: You rent the entire physical machine.

The difference between a VPS and a cloud server lies not in the technology but in how the service works. According to the AWS decision guide, Lightsail, its VPS product, comes with fixed monthly pricing and seven instance types, while EC2 offers pay-as-you-go or commitment-based pricing, hundreds of instance types and advanced networking options. Cloud providers rent out physical servers as well: an AWS Dedicated Host is a physical server fully dedicated to you, and it lets you use existing per-socket or per-core licenses such as Windows Server and SQL Server.

Pricing models: on-demand, committed and spot

Prices change often, so we describe the models, not the figures:

  • On-demand: No commitment and no upfront payment. AWS bills by the hour or by the second depending on the operating system (with a 60-second minimum); Azure charges only for the minutes used in a partial hour.
  • Committed: A discount in return for a 1- or 3-year commitment: Savings Plans and Reserved Instances on AWS, reservations on Azure. A commitment is a billing discount, so on AWS a Reserved Instance is billed until the end of its term even if the instance it applied to has been terminated.
  • Spot: The provider's spare capacity, priced below on-demand. When the provider needs the capacity back, the instance is interrupted; AWS gives two minutes' warning and Azure 30 seconds, and Azure offers no SLA for these VMs. Spot suits interruption-tolerant work such as batch jobs, analytics and testing.

Server hours are not the whole bill: disks, IP addresses and data transfer are separate line items. We cover the trade-off between paying as you go and investing upfront in our CAPEX vs OPEX article.

How to choose a cloud server

  • Workload profile: Is the bottleneck CPU, memory or disk? Azure recommends its memory optimized sizes for relational database servers. A steady load suits a commitment, a fluctuating load on-demand pricing, and interruption-tolerant work spot capacity.
  • Scaling: On AWS, moving to a larger instance type means stopping the instance. The way to add capacity without stopping anything is to raise the number of instances with Auto Scaling and spread traffic across them with a load balancer.
  • Region and data residency: AWS recommends choosing a region close to your customers or one that meets your legal requirements. If the server will process personal data, this is also a decision under the KVKK (Türkiye's personal data protection law); we cover the conditions for cross-border transfers in our KVKK article.
  • Availability: The EC2 SLA commits to 99.5% for a single instance and 99.99% for deployments spread across at least two zones in the same region. We explain how to read an SLA in our uptime article.
  • Backup: AWS replicates EBS data across multiple servers within a zone, but that copy will not bring back data deleted by mistake, because the deletion is replicated to every copy too. You need snapshots and a tested backup, and nothing is copied across regions automatically. More in our backup and disaster recovery articles.
  • Operations and support: As Microsoft points out, configuration, patching and software installation are still your job in the cloud. Who handles monitoring, backup testing and incident response should be settled up front, along with the scope of the provider's support. We describe ways to hand this work over on our managed cloud services and infrastructure management pages.

Frequently asked questions

What does cloud server mean? A virtual server that runs on a provider's infrastructure, is rented over the internet and is paid for as you use it.

What is a cloud server used for? It runs anything a server runs, such as websites, applications, databases and test systems, without buying hardware. Capacity can be changed as needs change.

What is the difference between a cloud server and a VPS? Both can be virtual servers; the difference is in how the service works. AWS's VPS product, Lightsail, comes with fixed monthly pricing, while EC2 offers hundreds of instance types, usage-based billing and automatic scaling.

How is a cloud server priced? There are three models: on-demand with no commitment, discounted pricing for a 1- or 3-year commitment, and spot capacity that can be interrupted. Disks, IP addresses and data transfer are charged separately.

Do charges stop when a cloud server is shut down? Compute charges stop, but disks and static IP addresses keep being billed. On Azure, a VM shut down from inside the operating system keeps incurring compute charges until it is deallocated.

Sources