Hosting glossary
The terms that appear on hosting specification sheets, defined in the way that matters when you are comparing plans.
- Bandwidth vs transfer
- Bandwidth is the rate of the pipe (Mbps, Gbps); transfer is the volume moved over a month (GB, TB). A plan can be generous in one and capped in the other, and the cap is usually the one you meet first.
- CDN
- A network of caching servers in many locations. The nearest one answers, so the distance a request travels — and the load on your origin — both fall.
- Control panel
- The web interface that manages accounts, domains, databases and mail on shared and reseller hosting. cPanel, Plesk and DirectAdmin are the common ones; the panel defines what you are permitted to do.
- ECC memory
- Memory that detects and corrects single-bit errors. Standard on server hardware and absent from the cheapest dedicated offers. For anything storing data, worth insisting on.
- Egress
- Data leaving a provider's network. Metered on cloud platforms, usually included on VPS and dedicated plans, and the most common cause of an unexpected cloud bill.
- Entry processes
- A shared-hosting limit on how many of your requests may execute simultaneously. Hitting it produces 503s while the server itself is healthy.
- Failure domain
- A set of things that fail together — a machine, a rack, a power feed, an availability zone. Redundancy means spanning domains, not duplicating within one.
- Hypervisor
- The layer that divides a physical machine into virtual ones. KVM, Xen and VMware are hypervisors; OpenVZ and LXC are container systems that share the host's kernel instead.
- Inodes
- A count of files and directories. Shared hosts cap it, and caches, mail and backups consume it quickly. Running out stops writes with plenty of disk free.
- IOPS
- Input/output operations per second — how many separate disk reads or writes the storage can perform. Databases care about this far more than about sequential throughput.
- IPMI / iDRAC / iLO
- Out-of-band management on server hardware: console access, power cycling and OS reinstallation independent of the operating system. The difference between a five-minute recovery and a support ticket.
- KVM
- Full virtualisation: your VPS gets its own kernel and can run anything, including Docker and custom modules. Distinct from container virtualisation.
- Managed vs unmanaged
- Who administers the operating system. Unmanaged means the provider keeps the hardware and network running and nothing else. The word "managed" is used loosely — ask exactly where the line falls.
- Oversubscription
- Selling more virtual resource than physically exists, on the assumption tenants do not all peak together. Universal, never published, and the main reason identical-looking plans perform differently.
- RAID
- Combining disks for redundancy, speed or both. RAID 1 mirrors, RAID 10 mirrors and stripes, RAID 5/6 use parity. None of them is a backup — they protect against disk failure, not against deletion.
- Snapshot
- A point-in-time image of a volume or instance. Fast to take and restore, and usually stored on the same platform — which makes it excellent for upgrades and insufficient as your only backup.
- Steal time
- CPU time your virtual machine wanted but did not get because the hypervisor gave it to someone else. Visible in `top` as `st`, and the clearest measurable symptom of a crowded host.
- TTFB
- Time to first byte — how long before the response starts arriving. Dominated by server work and distance, and the metric that caching and edge delivery improve most.
- Uptime SLA
- A contractual availability figure with a remedy attached, usually a service credit. An SLA without a credit is a marketing claim; a credit worth a few percent of one month's fee is not insurance against your own downtime.
- vCPU
- A virtual CPU presented to a guest — typically one hardware thread, shared or dedicated depending on the plan. Two vCPU on different platforms are not comparable quantities.
Each of the eight hosting models has its own guide if a definition raises a bigger question.