This is currently a work in progress and comments are welcome
- Glossary of Data Center and Networking Nomenclature
- Table of Contents
- I. Data Center Nomenclature
- II. Sizing, Metrics, and Density
- III. Edge and Distributed Infrastructure
- IV. Connectivity and Interconnection
- V. Facility Infrastructure and Operations
- VI. Construction and Form Factors
- VII. Security and Access
- VIII. Reliability and Standards Nomenclature
- IX. Internet Exchange Point (IXP) Operating Models
- Appendix: Summary of Data Center Scale Classifications
AI Factory (AI Data Center) A marketing term for a specialized facility purpose-built for the training and operation of massive artificial intelligence (AI) models, characterized by an architecture centered on Graphics Processing Units (GPUs). These facilities exhibit Extreme Density (often 40–100+ kW per rack), necessitating advanced liquid cooling. Unlike traditional enterprise data centers prioritizing absolute uptime (2N redundancy), AI Factories may accept reduced electrical redundancy (e.g., N+1 or N) to maximize raw power capacity and cooling efficiency, as workloads are generally check-pointed.
Colocation Data Center (Colo / MTDC) Third-party, multi-tenant facilities where the provider constructs and manages the physical infrastructure (power, cooling, security), and tenants lease space (racks, cages, suites) for their hardware. Enables organizations to outsource data center requirements, shifting from CAPEX to OPEX while gaining enterprise-grade infrastructure. Often utilized as a hub for interconnection and hybrid cloud architectures. See also: Multi-Tenant Data Center
Carrier Hotel (Telco Hotel) A specialized colocation data center typically in central urban locations, focused on telecommunications carriers and network connectivity. Their primary function is facilitating traffic exchange (Peering) via extensive fiber interconnections and Meet-Me-Rooms (MMRs). The core value proposition is Network Density. Often older legacy buildings with lower power density but massive ecosystem gravity.
Cryptocurrency Mines Industrial facilities engineered around the arbitrage between energy costs and the market value of mined coins. Operating under a "Minimal Viable Infrastructure" philosophy, they deliberately exclude redundancy (UPS, generators) because operational downtime is an acceptable business risk. Common cooling methods include Open-Air or evaporative cooling.
Data Center Provider The entity providing space, power, cooling, security, and related services to customers, including the IXP.
Enterprise Data Center A private facility owned, operated, and exclusively utilized by a single organization. Emphasizes risk mitigation, control, and customization to meet specific operational or compliance mandates. Capacities range from hundreds of kilowatts up to 20 MW; essential for regulated industries requiring Data Sovereignty.
Hyperscale Data Center Massive facilities designed and operated by technology conglomerates (AWS, Google, Microsoft, Meta) to support cloud services and vast compute/storage scalability. Defined by extreme scale (>5,000 servers, 100,000+ sq ft, 20 MW to 650+ MW). They prioritize extreme operational efficiency (PUE often ≤1.1) and utilize custom hardware (White-Box Servers).
Managed Services Data Center Facilities where a Managed Service Provider (MSP) offers comprehensive management solutions (server, storage, network administration, 24/7 monitoring), enabling complete infrastructure outsourcing.
Retail Colocation Colocation tailored to customers requiring smaller footprints. Space is typically leased by the rack, cage, or Rack Unit (U), generally fewer than 10 racks or <100kW. Frequently incorporates "Remote Hands" support.
Wholesale Colocation Leasing large blocks of power and dedicated space (entire halls or floors) to a single tenant. Pricing is based on Megawatts (MW) and square footage, usually metered. Typical leases are 5–15 years for 1–5 MW increments.
Supercomputing Centers (HPC) Government, academic, or research facilities designed for high-end scientific modeling and simulation. Performance is measured in Flops (Floating Point Operations per Second), utilizing highly customized computational hardware.
Mega Data Center The largest classification based on rack count, accommodating 9,000+ racks with power demands typically exceeding 100MW, serving massive cloud operations.
Multi-Tenant Data Center (MTDC) Synonym for Colocation Data Center
Data Center Campus A regionally local cluster of Data Centers operated by a single entity, allowing for seamless interconnection between tenants across different facilities.
Critical IT Load The definitive unit of measurement adopted by the industry for data center sizing, supplanting square footage.
Kilowatt (kW) The standard unit for measuring Power Density or power consumption of individual racks. Utilized for sizing smaller facilities (Micro/Enterprise).
Megawatt (MW) The standard unit for quantifying the aggregate capacity of Wholesale, Colocation, and Hyperscale facilities. Refers to the conditioned power a facility can continuously supply to the IT floor.
Raised Floor Space The historical primary metric for measuring data centers (sq ft or sq m). Due to increasing rack density, it is now considered a secondary or misleading metric for technical sizing.
Power Density The power draw per rack, expressed in kW/rack, which determines the required cooling architecture.
-
Low Density: <5 kW/rack (Legacy enterprise).
-
Standard / Medium Density: 5–10 kW/rack (Modern retail colo / general cloud).
-
High Density: 10–30 kW/rack (Intensive virtualization / storage arrays).
-
Ultra-High / Extreme Density / AI-Ready: >30 kW/rack. Air cooling becomes infeasible, necessitating liquid cooling.
Liquid Cooling A thermal management technique that replaces or augments air cooling by using fluids (water, dielectric fluids) to remove heat from IT components. Essential for Ultra-High Density racks (AI/HPC) exceeding 30 kW, where air cooling is physically insufficient. Approaches include Direct-to-Chip (DtC) and Immersion Cooling.
Power Usage Effectiveness (PUE) A metric assessing energy efficiency (Total Facility Power / IT Equipment Power). A PUE of 1.1 or lower signifies exceptionally high efficiency.
Flops Floating Point Operations per Second; the performance metric used in Supercomputing.
Petascale Data Center Facilities capable of 10^15 operations per second.
Exascale Data Center Facilities achieving 10^18 operations per second.
Edge Computing A computational topology emphasizing the physical placement of compute and storage resources in close proximity to the user or data generation point to mitigate latency, optimize bandwidth, and enhance reliability.
Edge Data Center Decentralized physical facilities strategically situated near end-users or primary data sources. Essential for ultra-low latency applications (5G, IoT, autonomous vehicles). IT load capacity typically ranges from 50 kW to 500 kW.
-
Extra Small: Extra Small Edge Facility - Street Furniture or Pole/Wall Mounted System.
-
Small: Small Edge Facility. Entry Level. Small Cabinet System or Street Furniture 0-4 Racks.
-
Medium: Medium Edge Facility. 4-14 Racks. Basic Redundancy, Low to Moderate Density.
-
Large: Large Edge Facility. 8-24 Cabinets/Racks. Moderate Redundancy or High Density.
-
Extra Large: XL Edge Facility. 24+ Cabinets/Racks. High Density. High Redundancy.
Regional Edge / Metro Edge Installations in "Tier 2" metropolitan areas or suburban peripheries of major hubs. Function as localized distribution/caching points. Scale generally ranges from 100 kW to 10 MW.
Micro Data Center (µDC) Highly compact, self-contained environments integrating all infrastructure (power, cooling, fire suppression) within a condensed footprint (often a single enclosure). Capacity spans 50 kW to 150 kW (1 to 10 racks).
Nano Data Center The extreme atomization of infrastructure. Direct integration of computing power into Consumer Premises Equipment (CPE). Often a single rack (24U-42U) drawing <3.5-7kW, or smaller integrated devices.
Network Edge / Far Edge / Tower Edge Infrastructure located at the physical boundary of a telecom operator's network (cell towers, utility cabinets) for sub-5ms latency processing.
Internet Exchange (IX) An IX is the entity that operates one or more switching fabrics at one or more locations for the purpose of peering (interconnection).
Internet Exchange Point (IXP) A network facility that enables the interconnection of more than two independent Autonomous Systems, primarily for the purpose of facilitating the exchange of Internet traffic. An IXP provides interconnection only for Autonomous Systems. An IXP does not require the Internet traffic passing between any pair of participating Autonomous Systems to pass through any third Autonomous System, nor does it alter or otherwise interfere with such traffic. Note that “Autonomous Systems” has the meaning given in BCP6/RFC4271.
Cable Landing Station (CLS) Coastal facilities where submarine cables terminate and connect with terrestrial networks. Often integrated with data centers.
Carrier Neutral A facility (or MMR) permitting any carrier to provide services, preventing operator monopolies on bandwidth and lowering costs for tenants.
Meet-Me-Room (MMR) The central physical nexus of a highly connected facility. A secure, managed area where fiber optic cables from various tenants and carriers are cross-connected. Should ideally contain little to no powered equipment.
MMR Operator The entity that controls and operates one or multiple designated Meet-Me-Rooms (MMRs) in the building. The MMR operator provides the infrastructure and management to support interconnection of network providers with each other and cross-connection either from network provider to building tenant or amongst building tenants. The MMR may include a variety of media types, including but not limited to: single and multi-mode fiber as well as Cat 5 copper or similar legacy infrastructure.
Cross Connect A physical patch cable (usually fiber) installed to connect one tenant's equipment to another's (or to a carrier). Can be routed via an MMR or run direct (Home Run) from rack to rack.
Virtual Cross-Connect A software-defined interconnection that logically functions like a physical cross-connect. Often provisioned instantly over a shared network fabric or Software-Defined Network (SDN) platform.
Data Gravity The phenomenon where a large aggregation of data attracts additional applications and services to co-locate nearby to minimize latency and transfer costs.
Interconnection Hub A modern designation for a highly connected facility accommodating carriers, Cloud On-Ramps, CDNs, and IXPs.
Peering (Paid vs. Settlement-Free) The reciprocal exchange of traffic between distinct networks.
-
Settlement-Free Peering: Networks exchange traffic without exchanging money, assuming mutual benefit.
-
Paid Peering: One network pays the other for access to their network routing table, often bordering on transit agreements.
Cloud Onramp A dedicated, private physical or virtual connection directly into a public cloud provider's edge node (e.g., AWS Direct Connect, Azure ExpressRoute). It bypasses the public internet for better security and predictable latency.
Backhaul The portion of a hierarchical telecommunications network linking the core network (backbone) to the edge subnetworks.
Last-Mile The final leg of the telecommunications network delivering connectivity directly to the retail end-user or enterprise premise.
Metro Interconnect High-capacity network fiber connections spanning a metropolitan area, linking distinct data centers, carrier hotels, and enterprise offices.
Data Center Interconnect (DCI) Networking solutions utilizing high-speed packet-optical connectivity to link two or more data centers over short, medium, or long distances.
Layer 2 vs. Layer 3 Interconnect
-
Layer 2 Interconnect: Connects networks at the Data Link layer (MAC addresses, VLANs). Used heavily in IXPs to allow peering across a shared broadcast domain.
-
Layer 3 Interconnect: Connects networks at the Network layer via routing protocols (IP addresses, BGP, OSPF).
VxLAN (Virtual Extensible LAN) A network virtualization technology that encapsulates MAC-based Layer 2 frames within Layer 4 UDP packets. Crucial for massive cloud computing, allowing Layer 2 overlay networks to scale across Layer 3 boundaries.
MPLS (Multiprotocol Label Switching) A routing technique that directs data from one node to the next based on short path labels rather than complex network addresses, speeding up traffic flow and allowing for traffic engineering.
Edge Router A specialized, high-capacity router situated at the boundary of a network, handling traffic leaving or entering the autonomous system.
Facilities-based Network Provider The entity that controls and maintains a physical fiber path into a building. Multiple providers may lease or share this fiber.
LACP / LAG (Physical Interface Speeds) Link Aggregation Group (LAG) combines multiple physical network ports into a single logical port to increase bandwidth and provide redundancy. Link Aggregation Control Protocol (LACP) is the standard protocol that negotiates and manages this bundling.
Fiber Distribution Panel (FDP) / Optical Distribution Frame (ODF) High-density, passive physical infrastructure where external or campus fiber cables are spliced, organized, and cross-connected to internal equipment.
Optical Splitter A passive component that takes a single beam of light and divides it into multiple outbound beams (e.g., a 1:32 or 1:64 split). It is the core enabler of all Passive Optical Network (PON) architectures.
GPON (Gigabit Passive Optical Network) A point-to-multipoint access standard utilizing passive splitters to provide high bandwidth (typically 2.5 Gbps downstream, 1.25 Gbps upstream) over a single fiber strand.
OLT (Optical Line Terminal) The provider-side aggregation equipment of a GPON network. It sits in the data center or central office, controls bandwidth allocation, and manages downstream endpoints.
ONT / ONU (Optical Network Terminal / Unit) The end-user boundary device that converts optical signals back into usable electrical signals (Ethernet, Wi-Fi). ONT and ONU refer to essentially the same hardware at the customer premise, depending on the specific standard body referenced.
ODN (Optical Distribution Network) The entirely passive physical fiber infrastructure (the cables, splitters, and FDPs) residing between the OLT and the ONT/ONU.
Wavelength Division Multiplexing (WDM/DWDM) Optical multiplexing technology that multiplexes multiple optical carrier signals onto a single optical fiber by using different wavelengths (colors) of laser light, vastly increasing total network capacity.
Optical Transceivers (SFP, QSFP, OSFP) Hot-pluggable optic modules that interface with edge routers and switches, converting electrical data signals into light for transmission over fiber optic cables.
IXP Member, Participant, and Customer Terminology
-
Participant: Broadly, any entity physically and logically connected to an IXP switching fabric.
-
Member: An entity that participates in an IXP, often implying governance, voting rights, or cooperative ownership (especially in non-profit/community IXPs).
-
Customer: An entity paying for access to an IXP, a term usually favored by For-Profit IXP operators.
-
Member Network: The actual Autonomous System (AS) and physical routing equipment terminating the connection.
-
IX End User: The consumer or enterprise downstream from the peering networks, whose traffic is improved by the IXP.
Medium Voltage (MV) Switchgear Heavy-duty electrical equipment used to control, protect, and isolate power systems. Bridging utility transmission and site distribution, medium voltage in data centers generally falls between 15kV and 38kV.
Automatic Transfer Switch (ATS) A critical mechanism that automatically transfers the facility's electrical load from the primary utility feed to the backup generators in the event of a power anomaly or outage.
Uninterruptible Power Supply (UPS) A critical electrical apparatus that provides instantaneous emergency power to a load when the primary input power source fails, typically using batteries or flywheels. It bridges the gap until standby diesel generators can start and synchronize.
Power Distribution Unit (PDU)
-
Floor PDU / Remote Power Panel (RPP): Standalone cabinets downstream from the UPS that step down facility voltage to usable rack voltage and distribute it via circuit breakers to the data hall rows.
-
Rack PDU (rPDU): The power strip mounted directly inside the IT cabinet, ranging from basic distribution to fully metered, remotely switched models.
CRAC / CRAH
-
CRAC (Computer Room Air Conditioner): Uses a built-in compressor and chemical refrigerants to cool the air.
-
CRAH (Computer Room Air Handler): Uses fans and cooling coils through which chilled water (supplied by a central plant) circulates to remove heat.
Hot Aisle / Cold Aisle A fundamental data center layout design. Server racks are aligned in alternating rows so that cold air intake faces one way (Cold Aisle) and hot air exhaust faces the other (Hot Aisle). This prevents the mixing of hot and cold air, drastically improving cooling efficiency.
Central Utility Plant (CUP) A centralized facility separated from the IT data halls that houses the primary, heavy infrastructure for producing chilled water (chillers, cooling towers) and sometimes primary electrical switchgear.
NOC (Network Operations Center) A centralized location where engineers continuously monitor, manage, and maintain the health and performance of the network infrastructure and switching fabrics.
FOC (Facility Operations Center) The command center dedicated to monitoring and managing the physical infrastructure of the data center, including power distribution, cooling, environmental sensors, and physical security.
SOC (Security Operations Center) A centralized unit that monitors an organization's IT infrastructure for cybersecurity threats, defending against intrusions and managing incident response.
Base Building Refers to the entity holding fee simple ownership or primary control of the underlying real estate. While this entity retains ultimate authority over network entry rights, it may delegate broad access privileges to an MMR Operator or Data Center Provider.
Brownfield Data Center Existing structures retrofitted for data center use. Facilitates expedited deployment but may encounter infrastructure limitations.
Greenfield Data Center Entirely new facilities constructed on undeveloped land, designed to precise specifications without pre-existing constraints.
Containerized Data Center Infrastructure integrated within standard ISO shipping containers (20ft or 40ft). Highly mobile, suitable for disaster recovery or remote capacity (typically 13–26 racks).
Mobile Data Center Fully portable centers mounted on trucks or trailers, enabling rapid deployment within hours.
Modular Data Center (MDC) Systems assembled using pre-engineered, prefabricated modules assembled and tested in a factory. Reduces deployment timelines significantly.
Prefabricated / Skid-Mounted Data Center Critical components (Power/Cooling skids) pre-assembled on structural steel frames in a factory, engineered for swift installation within a shell building.
Powered Shell Data Center Partially completed facilities where the developer finalizes the building exterior and core utilities, allowing tenants to customize the interior build-out.
Underground / Floating Data Centers
-
Underground: Built in mines/bunkers; provides physical hardening and natural thermal regulation.
-
Floating (Data Barges): Situated on vessels leveraging surrounding water for highly efficient heat rejection.
Mantrap (Security Portal) A physical security access control system consisting of a small space with two sets of interlocking doors. The first door must close and lock before the second door can be opened, preventing tailgating and ensuring proper biometric/badge authentication.
Anti-Tailgating / Anti-Passback
-
Anti-tailgating: Sensor or turnstile systems designed to prevent an unauthorized individual from following an authorized person through a secure doorway.
-
Anti-passback: System programming that prevents a credential holder from passing their badge backward through a door for another person to use.
Biometric Access Control The use of unique physical characteristics—such as fingerprint, iris, or facial recognition—to verify identity. This ensures physical presence rather than mere possession of a keycard.
Multi-Factor Authentication (MFA) Requiring two or more independent forms of verification to grant physical access, typically combining something a person has (RFID badge), something they know (PIN), and something they are (biometrics).
Defense-in-Depth (Physical) A layered security architecture creating multiple, concentric rings of protection (e.g., perimeter fencing -> lobby turnstiles -> data hall mantrap -> locked server cabinet). A breach of one layer does not compromise the entire facility.
Bollards (Crash-Rated) Heavy-duty steel or concrete perimeter barriers specifically engineered to protect the facility's exterior walls, entryways, and external generators from vehicle-ramming attacks.
Role-Based Access Control (RBAC) In a physical security context, this involves restricting an employee or vendor's badge access strictly to the specific data halls, cages, or zones required for their job function, rather than granting blanket facility access.
Uptime Institute Tiers The international standard for classifying data center availability:
-
Tier I (Basic Capacity): Single path, no redundancy (N). 99.671% availability.
-
Tier II (Redundant Capacity): Single path, redundant components (N+1). 99.741% availability.
-
Tier III (Concurrently Maintainable): Multiple paths, N+1 redundancy. Any component can be removed without shutdown. 99.982% availability.
-
Tier IV (Fault Tolerant): Fully redundant (2N or 2N+1), multiple active paths. Withstands a single fault without impacting IT load. 99.995% availability.
TIA-942 Ratings A standard by the Telecommunications Industry Association ("Rated-1" through "Rated-4"). Broader scope than Uptime, covering cabling, architecture, and physical security.
BICSI Availability Classes Standard ranging from Classes 0 through 4. Classes 0 and 1 address facilities where downtime is acceptable (e.g., crypto mining).
Availability Zone (AZ) Logical groupings of proximate data centers within a Cloud Region. Each AZ operates with independent power/cooling/networking to ensure fault tolerance but is treated logically as a single failure domain by cloud platforms.
(Note: Definitions sourced from "Community-Driven IXPs: Enhancing Local Connectivity and Sustainability" by Internet Society, CC BY-NC-SA 4.0)
Member-Operated IXP (Community-driven) Governance is led by members through an elected board. Non-profit, democratic organizations created to enhance connectivity for their members.
ISP Association-Operated IXP Established by a consortium of major ISPs who collaborate on bylaws. Staffing is typically funded by the consortium to optimize regional interconnectivity.
For-Profit IXP A commercial venture managed by private entities generating revenue by offering a neutral peering environment and a broad range of services.
Academia/Regulator-Managed IXP Management overseen by an academic institution or regulator. Supports research and education goals while functioning as a neutral peering point.
Informally Managed IXP Lacks formal structure; relies on mutual cooperation of benefiting networks. Long-term viability is often debated.
ISP-Operated IXP Run by a single ISP. Can create neutrality concerns; may require regulatory oversight to ensure fair access.
| Classification | Typical Physical Footprint | Typical Power Capacity | Key Application |
|---|---|---|---|
| Nano | 1–50 sq ft (1 Rack/Enclosure) | < 7 kW | IoT Gateways, CPE integration, Fog computing |
| Micro (µDC) | 1–500 sq ft (1–10 Racks) | < 100 kW – 150 kW | Cell Tower installations, Smart Buildings |
| Mini | Up to 10 racks | Basic capacity | Supporting startups/small businesses |
| Small / Server Room | 1,000–5,000 sq ft | 100 kW – 500 kW | SMB on-premises servers |
| Medium / Enterprise | 5,000–20,000 sq ft | 1 MW – 5 MW | Corporate HQ, Private Cloud, Regional Hubs |
| Regional Edge | 50,000–100,000 sq ft | 2 MW – 10 MW | Localized content delivery caching |
| Large / Wholesale | 20,000–100,000 sq ft | 5 MW – 20 MW | Multi-tenant Colo, Enterprise Hubs |
| Massive / Hyperscale | 100,000+ sq ft | > 20 MW (up to 650+ MW) | Cloud Service Providers, Major Social Media |
| Mega Campus | 1M+ sq ft (multi-building) | > 100 MW (up to 240+ MW) | Largest cloud service operations |