A server is a computer system, or a program running on one, that provides services, data or computing resources to other devices, called clients, over a network. The term describes a role in the client-server model as much as a type of machine: the same physical computer can run a web server, a database server and a DNS server at once, while a laptop becomes a server the moment it answers network requests.
In server fundamentals, the word usually refers to server-class hardware and the operating system and applications running on it, designed for continuous, unattended operation. What distinguishes such a machine from a desktop is not raw speed but reliability, remote manageability and the ability to serve many clients concurrently.
How a server works in the client-server model
A server application listens on a network port and waits for requests. A client opens a connection, sends a request in an agreed protocol such as HTTP, SMTP or SQL over TCP, and the server processes it and returns a response. A web server conventionally listens on TCP port 443 for HTTPS, while a DNS server answers on port 53.
To serve thousands of clients at once, server software uses concurrency: multiple processes, threads or event-driven loops that handle many connections in parallel. The operating system provides the network stack, scheduling, file systems and security isolation, and the hardware supplies processors, memory, storage and network interfaces sized for the expected load.
Servers are typically headless, with no monitor or keyboard attached, and are administered remotely over SSH, remote desktop protocols or a dedicated management controller. They run continuously, often for months between planned restarts, in a server room or data center with controlled power and cooling.
What makes server hardware different
Server-class hardware is built around reliability, availability and serviceability. Each design choice reduces the chance that a single component failure stops the service or that a technician must take the machine offline for routine work.
| Feature | Server-class hardware | Typical desktop | Why it matters |
|---|---|---|---|
| Memory | ECC registered memory, often hundreds of gigabytes | Non-ECC, 8 to 64 GB | Detects and corrects single-bit errors |
| Power supply | Redundant, hot-swappable units | Single unit | Survives a power supply failure |
| Storage | Hot-swap drive bays with RAID | Fixed internal drives | Replaces failed drives without downtime |
| Remote management | Baseboard management controller | None or limited | Console, power and sensors over the network |
| Processors | Many cores, often two or more sockets | One socket, fewer cores | Handles many concurrent workloads |
| Duty cycle | Designed for continuous operation | Designed for working hours | Components rated for constant load |
The baseboard management controller (BMC) deserves special mention. It is an independent microcontroller with its own network port that allows administrators to power the server on or off, view the console and read temperature and fan sensors even when the operating system has crashed, using interfaces such as IPMI or the newer Redfish API.
Server form factors
Physical servers come in three main form factors, which differ in density, cooling and where they are installed. Rack heights are measured in rack units (U), where 1U equals 1.75 inches or 44.45 mm, and a standard full-height rack provides 42U of mounting space.
| Form factor | Characteristics | Typical setting |
|---|---|---|
| Tower server | Freestanding case, quiet, easy to expand | Small offices without a server room |
| Rack server | Mounts in a 19-inch rack, usually 1U or 2U high | Data centers and server rooms |
| Blade server | Thin modules sharing power, cooling and networking in a chassis | High-density enterprise data centers |
| Edge server | Compact, ruggedized, tolerant of heat and dust | Factories, retail stores, telecom sites |
Physical, virtual and cloud servers
A server today is often a software construct rather than a box. A hypervisor can divide one physical machine into many virtual machines, each running its own operating system and behaving as an independent server. This raises hardware utilization but introduces shared resources and an additional software layer.
| Type | What the customer controls | Main trade-off |
|---|---|---|
| Owned physical server | Everything from hardware to applications | Full control, but capital cost and maintenance |
| Dedicated or bare metal server | An entire rented physical machine | No neighbors, but slower to provision and scale |
| VPS | A virtual machine on shared hardware | Low cost, but resources shared with other tenants |
| Cloud instance | A virtual or bare metal machine billed by usage | Elastic and fast to provision, but costs grow with usage |
| Container | An isolated process sharing the host kernel | Very lightweight, but weaker isolation than a virtual machine |
Whatever the type, a server goes through server provisioning: allocating hardware or a virtual instance, installing the operating system, applying configuration and security settings and deploying applications. Automating this process is what allows modern infrastructure to replace a failed server in minutes rather than days.
Availability, sizing and common mistakes
Server reliability is usually expressed as availability, the percentage of time a service is usable. Each additional nine reduces allowed downtime tenfold, and reaching the higher levels requires redundant servers and automatic failover, because no single machine can guarantee them.
| Availability | Maximum downtime per year | Typical requirement |
|---|---|---|
| 99% | About 3.65 days | Internal tools, test systems |
| 99.9% | About 8.76 hours | Standard business applications |
| 99.99% | About 52.6 minutes | E-commerce and customer-facing services |
| 99.999% | About 5.26 minutes | Telecom and critical infrastructure |
- Treating one server as the whole service leaves no path for maintenance or failure; critical services need at least two instances.
- Sizing for average load instead of peak load leads to slow responses exactly when demand is highest.
- Exposing management interfaces such as the BMC to the internet creates a serious attack path; they belong on an isolated management network.
- Skipping backups because of RAID confuses redundancy with recovery; RAID does not protect against deletion, corruption or ransomware.
The short version
A server is a computer whose job is to answer requests from other computers. When a phone opens a website or an office laptop saves a file to a shared drive, a server somewhere receives that request and sends back what was asked for.
It works like a library service desk that never closes. Visitors ask for something, staff find it and hand it over, and the desk must stay open for many people at once. That is why servers are built to run day and night, to keep working when one part fails and to be managed from a distance.
Where it makes a difference
A medical practice with 25 staff stores patient records, scanned documents and its scheduling database on a single desktop computer under the reception desk. When that computer’s only hard drive fails on a Monday morning, the practice loses access to appointments for most of the day while files are restored from a week-old backup.
The practice replaces it with a 2U rack server in a small locked cabinet, fitted with ECC memory, redundant power supplies and four drives in RAID 10 so that any single drive can fail without interruption. An uninterruptible power supply bridges short outages, nightly backups go to a separate device and an off-site copy, and the BMC sends an alert when a drive or fan reports a fault.
Two years later, a drive fails during clinic hours. The server keeps running, the alert reaches the IT contractor, and the drive is swapped while the system stays online, with no downtime and no data loss.