Tech, ExplainedNetworking, infrastructure and security
Product

Servers

Built to run all day and survive failure.

BEFORE ITAFTER ITServersDell Technologies (PowerEdge)HPELenovoPARTBaseboard managementcontroller (BMC)+PARTServer fleet management+PARTHardware root of trust+COVERSRedundant partsHot-swappingECC memoryRemote managementLoad balancers↗ ENTERPRISE NETWORKINGsendsuser requestsData centrepower and coolingsuppliespowerCPU↗ AI AND DATA CENTRE CHIPSrunsthe computingreads andwrites dataEnterprise storage systemssends trafficData centreswitches and routersruns virtualmachinesHypervisorand virtualisation↗ INFRASTRUCTURE SOFTWARECONCEPTS IT USESFirmwareBackendBEFORE ITAFTER ITServersDell Technologies (PowerEdge)HPELenovoPARTBaseboard managementcontroller (BMC)+PARTServerfleet management+PARTHardware root of trust+COVERSRedundantpartsHot-swappingECC memoryRemotemanagementLoad balancers↗ ENTERPRISENETWORKINGsendsuser requestsData centre powerand coolingsuppliespowerCPU↗ AI AND DATACENTRE CHIPSrunsthe computingreads andwrites dataEnterprisestorage systemssends trafficData centreswitchesand routersruns virtualmachinesHypervisor andvirtualisation↗ INFRASTRUCTURESOFTWARECONCEPTS IT USESFirmwareBackendBEFORE ITLoad balancers↗ ENTERPRISE NETWORKING↓ sends user requestsData centre power and cooling↓ supplies powerCPU↗ AI AND DATA CENTRE CHIPS↓ runs the computingServersDell Technologies (PowerEdge)HPELenovoBaseboard managementcontroller (BMC)+Server fleet management+Hardware root of trust+Redundant partsHot-swappingECC memoryRemote managementAFTER IT↓ reads and writes dataEnterprise storage systems↓ sends trafficData centre switches and routers↓ runs virtual machinesHypervisor and virtualisation↗ INFRASTRUCTURE SOFTWARECONCEPTS IT USESFirmwareBackend
The server receives user requests from the load balancer, power from the data centre and computing from its CPUs, and reads and writes data on storage, sends traffic over the switches and runs virtual machines.
opens a page inside this one a page in the same scene a page in another scene a concept page a label only, with no page

Servers are computers built to run all day, every day, and to serve many users at once. A server is made of the same basic parts as a laptop: a processor, memory, disks and a network connection. But it has many more of each part, and it keeps working even when one of them fails. Most servers sit in racks in a data centre.

How is a server different from a laptop? Take a typical rack server that is 2U high, which means it fills two slots of a rack, about 9 cm. It may have two CPUs with 32 or more cores each, while a laptop has one CPU with 8 to 16 cores. It may have 512 GB of memory, against 16 GB in a laptop. Its memory uses error correction (ECC), which finds and corrects a wrong bit before it can damage data. It has two power supplies and several fans, and each of them can be pulled out and replaced while the server keeps running. This is called hot-swapping. It also has fast network ports of 25 or 100 gigabits per second.

Imagine an online coaching company in Kota on the morning it publishes mock-test results. At 10 a.m., 2 lakh students log in within a few minutes. A load balancer spreads their requests across 40 servers. That afternoon, one power supply in one server fails. The server keeps running on its second power supply, and no student notices. Its baseboard management controller sends an alert, and an engineer replaces the failed unit the next morning without switching the server off.

What runs on a server? First, the firmware starts the hardware. Then an operating system starts, often an enterprise Linux. Many servers run a hypervisor, software that splits one physical server into many virtual machines, each with its own operating system. A server keeps its data on its own disks or on enterprise storage systems. Data centre switches join the servers to each other and to the internet.

Servers come in a few shapes. Rack servers are the most common. Blade servers are thin boards that slide into a shared box, called a chassis, which supplies their power and network. GPU servers hold 4 or 8 GPUs each, for AI work.

A few large companies make most of the world's servers. Their engineering centres in Bengaluru and Pune work mostly on firmware and management software. The very largest cloud providers design their own servers and have them built by contract manufacturers.

Tech, Explained is a part of ingrid.fyi. Every page explains one piece of technology in plain words.