A clear guide for decision-makers and businesses. Every term, no jargon.
Not an expert? That's fine. In a few minutes, here is all the vocabulary of datacenter hosting. Tap an icon to jump straight to a concept.
One U = 4.45 cm of height in a rack. It measures the size of equipment: a small server is often 1U, a bigger one 2U or 4U. You rent space in number of U.
💡 In practice: "I want 1U" = renting the slot of one thin server, the cheapest way to start.
A rack is the standard metal cabinet (often 42 to 48 U) that houses servers. You can rent a full rack, a half, a quarter, or just a few U. It's locked and powered.
💡 Choosing between 1U and a full rack means balancing cost and room to grow.
This is the secured space of the datacenter where racks are lined up in rows. Temperature and humidity are constantly controlled, and access is monitored (badge, cameras, biometrics).
To cool efficiently, the cold air is contained in a closed aisle facing the servers; the hot exhaust air is evacuated separately. Result: less wasted energy, so a lower electricity bill.
A datacenter must never shut down. On an outage: the UPS take over within milliseconds, then the diesel generators start. Redundancy is rated N+1 (one spare) or 2N (everything duplicated).
💡 "2N" = no possible outage even if a whole circuit fails. That's what guarantees uptime.
The PUE measures energy efficiency: it's the ratio of total power consumed to the power actually used by servers. 1.0 = perfect; a good modern datacenter is around 1.2–1.4.
💡 A low PUE = less waste = lower cost and reduced carbon footprint.
They prove, via an independent audit, that a datacenter delivers on its promises. The main ones:
Uptime Tier I→IV: availability/redundancy level. ISO 27001: information security. HDS: health-data hosting (France). ISO 22301: business continuity. SecNumCloud (ANSSI): trusted cloud.
💡 Why they exist: to reassure you without visiting — a third party has checked.
This is your access to the whole Internet: a carrier "transports" you to the rest of the world. It's billed by bandwidth used (often the 95th percentile) or as a flat fee per port (1G, 10G, 100G…).
💡 Without IP transit, your servers are not reachable from the Internet.
An IP pool (or block) is a set of Internet addresses assigned to you (e.g. a /29 = 8 addresses). Each service exposed on the Internet needs at least one public IP address.
An IXP is a hub where hundreds of carriers and companies interconnect directly (e.g. France-IX). Being in a datacenter linked to an IXP means shorter, faster and cheaper connections.
💡 The more connected a datacenter is (networks, IXP), the more attractive and competitive it is.
It's the carriers' meeting room at the heart of the datacenter. All networks arrive there; that's where your rack is linked to the carrier of your choice via a cross-connect.
It's the connection board where all cables arrive. Instead of running a long cable across the datacenter, links are "patched" cleanly through this panel. Essential for a tidy, scalable infrastructure.
Dark fiber is an optical fiber rented bare, with no equipment on it: you "light it up" with your own gear and do whatever you want. It's the highest-performance way to link two sites (near-unlimited capacity, low latency).
💡 Ideal to link two datacenters in a business-continuity plan (BCP/DRP).
DWDM sends several "colors" of light down a single fiber, each carrying an independent stream. It multiplies a fiber's capacity without laying new ones — perfect for large needs between sites.
It's the physical cable (usually a single-mode optical fiber) that links your rack to another party in the datacenter — a carrier, a cloud, a customer — through the Meet-Me Room. Billed as a one-off install + a monthly fee.
💡 It's the "socket" that physically connects you to a partner in the same building, without going over the Internet.
A PoP is where a carrier is physically present in a datacenter (its equipment is installed there). If it has a PoP in your building, it can easily connect you and offer links to its other sites.
💡 On DataColo, a partner declares its PoPs to offer inter-site links (useful for BCP/DRP).
The PDU is the rack's professional "power strip": it distributes electricity to the servers. Datacenters often provide two PDUs (on two circuits, A and B) for redundancy.
You rent a subscribed power capacity, expressed in kVA (or kW). The more your servers consume, the more kVA you need. It's often the 2nd cost item after space — billed inside the rack (flat) or on top (per kVA/month).
💡 Estimating your kVA well avoids paying for unused power.
Not on site? A datacenter technician acts on your behalf: reconnect a cable, reboot a server, receive equipment… Billed by the hour. Handy when you host far from home.
The SLA is the contractual uptime commitment: e.g. 99.99% = at most a few minutes of downtime per year. Below that, the datacenter owes penalties. The higher the SLA, the more redundant (and expensive) the infrastructure.
The BCP (Business Continuity Plan) keeps the service running during a disaster thanks to multiple sites; the DRP (Disaster Recovery Plan) restarts activity afterwards. Two key metrics: RPO (max tolerable data loss) and RTO (max recovery time).
💡 Two datacenters linked by fiber = the basis of a real recovery plan.
The TCO adds up all costs over time: space, electricity, IP transit, cross-connects, installation, handling, price escalation… DataColo computes it over 1, 3 and 5 years to compare datacenters on a fair basis.
💡 The rack price alone is misleading: it's the TCO that reveals the real bill.
In colocation, you own your servers and rent space + power + connectivity in a datacenter (costs controlled over time). In the cloud, you rent virtual resources on demand (flexible, but the bill grows with usage).
💡 For a stable, long-running workload, colocation is often much cheaper than the cloud.