Short answer

Traditional hosting and colocation usually begin with a machine. You choose or purchase a server with enough processing power, memory, storage and network capacity to run your applications. You then operate the software on it, or pay somebody to do that for you.

Cloud services let you separate those decisions.

An application can still run on a virtual server, but its files might come from object storage through a content delivery network. Incoming traffic can be distributed across several application instances. The database can be operated as a managed service. Certificates can be issued and renewed through the platform. Capacity can increase during busy periods and decrease when demand falls.

This does not make the cloud inherently better. It changes what can be selected, scaled and operated independently.

A dedicated server can be a perfectly sensible answer

A dedicated server is easy to understand. It has a known amount of processing power, memory, storage and network capacity. Its costs are generally predictable, and an experienced team has direct control over the operating system and installed software.

Colocation extends that control to the hardware itself. You choose the equipment, install it in a data centre and decide how it is connected and operated.

For a stable workload, either arrangement may work very well. If the application fits comfortably on one machine, demand is predictable and the necessary operational knowledge is available, moving it merely because cloud services exist would solve no obvious problem.

The limitations become more visible when the application changes.

A growing business may repeatedly replace its server with a more powerful one. More processors, more memory and faster storage can postpone the problem, but the application still depends on one machine. Maintenance affects everything running on it. A hardware problem has a large impact. Capacity must be selected for the busiest expected period, even if most of it remains unused at other times.

This is vertical scaling: making one machine larger. It is useful, but it cannot continue indefinitely.

Growth does not always happen evenly

Demand often changes by hour, day or season.

A business application may be busiest during office hours. An online shop may receive most of its traffic during a campaign or holiday season. A reporting process may need substantial processing power for half an hour and almost none for the rest of the day.

With purchased or dedicated hardware, capacity normally has to be available before that demand arrives. The server needed for the busiest week of the year may spend the other 51 weeks operating far below its limits.

Cloud platforms can make capacity easier to add and remove. That does not mean every application will scale automatically. An application built around one machine may store sessions, uploaded files or scheduled work locally. Starting a second copy could produce inconsistent data or duplicate processing.

The platform can provide additional capacity. The application must still be designed to use it.

A load balancer changes the growth model

Instead of sending every request to one increasingly powerful server, a load balancer can distribute traffic across several application instances.

This is horizontal scaling: adding more instances rather than continually enlarging one. Unhealthy instances can be removed from service, new ones can be added and capacity can be spread across more than one physical location. This can prevent the failure of one application instance—or even one physical location—from taking the complete service offline.

The arrangement works only when the instances are sufficiently interchangeable. Important state may need to move into shared storage or a database. Background work may need a queue. Deployments and configuration must be consistent. Monitoring must show what is happening across the complete service rather than on one machine.

A load balancer therefore does more than divide traffic. It introduces a different application and operating model.

Not every request needs to reach your application

A conventional web server may receive every request for every page, image, stylesheet and download. That is not always necessary.

Static files can be stored at an origin and delivered through a content delivery network, or CDN. The CDN keeps copies of suitable content at locations nearer to users. When a cached copy is available, the request does not need to travel to the original server.

This can reduce latency for users and reduce work at the origin. For a largely static website, the origin might be object storage rather than a continuously running web server. Interactive behaviour can still run in the browser or use separate APIs when needed.

Dynamic applications can also benefit. Images, scripts and other cacheable content may use the CDN while account functions, searches or transactions continue to reach the application.

The important distinction is not whether a website is interactive. It is which requests require processing at the origin.

A database does not have to mean another pair of servers

An important database creates operational work even when it is not particularly busy.

Running it yourself may involve installing and patching the database software, arranging backups, monitoring storage, configuring replication and preparing a secondary system for failure. The secondary machine consumes resources even when it performs little visible work.

Managed database services can take responsibility for parts of that work. Depending on the service selected, the provider may operate the underlying servers, install patches, create backups, monitor failures and maintain a standby instance.

The customer still has important responsibilities. A managed service does not design the data model, correct inefficient queries, decide who should have access or determine how much data loss the business can tolerate. A backup also has value only when it can restore the service within the required time.

The difference is that operating the database machinery can become a service choice rather than another collection of servers to build and maintain.

Some small jobs disappear into the platform

Cloud integration can remove work that used to require separate procedures.

A TLS certificate traditionally had to be requested, validated, received, installed and renewed. Cloud platforms can integrate certificates with DNS, load balancers and content delivery services. Once domain control has been established, issuance and renewal can often happen automatically—and, when everything works as intended, without anybody noticing that it happened.

The same principle applies elsewhere. Health checks can influence traffic routing. Logs can be collected without building a separate log server. Queues can connect application components without maintaining a message broker. Files can be retained according to storage policies rather than housekeeping scripts.

None of these jobs is impossible in a conventional environment. The difference is that they can be obtained as operated capabilities rather than assembled from software running on more machines.

Convenience creates dependency

The integrations that make cloud services convenient also make them harder to leave.

A virtual machine can often be recreated elsewhere with relatively familiar work. An application that depends on a provider’s identity system, database behaviour, queues, deployment tools, monitoring and access rules has a more complicated exit.

That dependency is not automatically a mistake. A managed service may save more time, reduce more risk or provide better availability than an organization could achieve independently. Avoiding every provider-specific feature can mean paying for the cloud while declining much of its practical value.

The dependency should nevertheless be understood before it accumulates.

Questions about portability are easier to answer while designing a service than during an urgent migration. Data volumes, export methods, network charges, replacement services and the knowledge available inside the organization all affect how realistic an exit would be.

Distance from the machinery can be an advantage

For people who enjoy building infrastructure, direct control has value of its own. Selecting hardware, installing racks, designing networks and arranging connectivity can be rewarding work.

For many organizations, the same activities are simply prerequisites for the application they actually need. Racks require space, power, cooling and physical security. Hardware ages. Cables fail. Network relationships and spare capacity have to be maintained.

Cloud services move some of that work away from the customer. This creates distance from the underlying technology, which can feel like a loss of control. It can also let a team spend more time on the service used by the business.

Whether that is an advantage depends on what the organization wants to operate itself.

Smaller parts can make later change easier

Moving an old server into the cloud without changing its structure may reproduce most of its existing limitations.

Separating functions can create more useful options. Static content, application processing, background jobs, databases and identity do not necessarily need the same scaling model or operational owner. A function that has clear boundaries can later be replaced without moving the complete environment with it.

That does not mean every application should be divided into dozens of microservices. Each separation introduces interfaces, permissions, monitoring and failure scenarios. A collection of small services can become harder to understand than one well-structured application.

The useful goal is not the maximum number of components. It is enough separation to let parts change independently where that independence has value.

The comparison starts with the workload

The practical question is not whether cloud services are better than dedicated servers or colocation.

It is where the current arrangement creates a constraint.

  • Are you continually replacing one server with a larger one?
  • Does capacity remain unused outside a seasonal or daily peak?
  • Would several application instances be able to share the work?
  • Is the team spending time operating databases, certificates or storage that a provider could manage?
  • Could frequently requested content be served closer to users?
  • Which parts require direct technical control?
  • What would happen if the chosen cloud service or provider had to be replaced?
  • Does the expected improvement justify the migration and the new dependencies?

Traditional hosting starts by asking what machine the application needs.

Cloud design can start by asking what each part of the application needs, how demand changes and who should operate it.

Sometimes the result is a cloud-native collection of services. Sometimes it is one virtual machine with a managed database and a CDN. Sometimes the existing dedicated server remains the best answer.

The benefit comes from having those choices—not from selecting the cloud in advance.

Sources and further reading

About the author

Harold Snippe

Harold Snippe is the founder of CS Infostructure. His experience spans building hosting and network infrastructure, operating international internet platforms and more than ten years of platform and security work within large banking environments.

About Harold and his background →

Questions, corrections and contributions

Talk to Harold

Have a question, a different experience or something that would make this article more useful? Leave a thought. Your feedback may help improve this page.

← Back to the Library