Direct Answer
This glossary covers the core vocabulary of enterprise connectivity, financial-market infrastructure and data-center interconnection—terms such as cross connect, colocation, dark fiber, cloud on-ramp and market data symbol—defined in plain language, with links to deeper explanations across the site.
Networking & connectivity
Cross connect
A direct physical cable connection between two parties’ equipment within the same data center or colocation facility. A cross connect avoids the public Internet and is commonly used to reach carriers, cloud on-ramps, exchanges and other tenants with a short, controlled in-building path. Learn more: Financial Markets connectivity →
Dark fiber
Unlit physical fiber-optic strands leased or purchased for exclusive use. The customer installs and operates the optical transport equipment, controls the upgrade path and assumes responsibility for lighting the fiber. Learn more: Ethernet vs. Wavelength vs. Dark Fiber →
DDoS mitigation
Network and security controls that detect and filter distributed denial-of-service traffic before it overwhelms a circuit, edge device or application. Learn more: Dedicated Internet and DDoS Protection →
Dedicated Internet Access (DIA)
A business-grade Internet service with committed bandwidth, provider support and service levels. DIA may also include BGP routing, static IP space and optional DDoS mitigation. Learn more: Dedicated Internet Access →
Ethernet service
A carrier-managed Layer 2 service used to connect locations, data centers or network environments with defined bandwidth and service levels. Learn more: Ethernet vs. Wavelength vs. Dark Fiber →
Jitter
Variation in latency over time. Low average latency with high jitter can create inconsistent application or trading performance. Learn more: Low-Latency Trading Networks →
LAN (Local Area Network)
The network inside a building, campus or local environment that connects users, devices, wireless access points and local systems. Learn more: Enterprise Connectivity →
Last mile
The final segment of a network connection between a provider’s network and the customer’s building. Last-mile availability, construction and building entry often determine installation timelines and route diversity.
Related: Route diversity, Ethernet service, dark fiber
Latency
The time required for data to travel between two points. It may be measured one way or round trip, so the test method and endpoints must be stated. Learn more: Low-Latency Trading Networks →
MPLS (Multiprotocol Label Switching)
A carrier-managed technology that forwards traffic across a private network using labels. MPLS is commonly used to connect multiple business locations and can support traffic prioritization and predictable service levels. Learn more: Building a WAN →
Network diversity
Resiliency designed across physical routes, carriers, facilities, entrances, equipment, power and upstream infrastructure rather than circuit count alone. Learn more: Network Diversity →
On-ramp (cloud on-ramp)
A private connection point into a cloud provider’s network, such as AWS Direct Connect or Azure ExpressRoute, used to reach cloud resources without relying solely on the public Internet. Learn more: Cloud Connectivity →
Private line
A dedicated point-to-point service reserved for one customer’s traffic between two locations. The service is private at the logical level, although underlying conduit or carrier infrastructure may still be shared. Learn more: Enterprise Connectivity →
Round-trip time (RTT)
The elapsed time for data to travel from an origin to a destination and for a response to return. RTT is different from one-way latency. Learn more: Evaluating ULL Connectivity →
Route diversity
Physical separation between network paths so that one likely fiber cut, conduit failure, facility event or equipment problem does not affect every service. Learn more: Network Diversity →
SASE (Secure Access Service Edge)
A network architecture that combines SD-WAN with cloud-delivered security services — such as secure web gateways, zero-trust access and firewall-as-a-service — so users and sites connect securely wherever they are.
Related: SD-WAN, Dedicated Internet Access (DIA), WAN
SD-WAN (Software-Defined Wide Area Network)
A software-controlled network layer that applies policy and dynamically directs traffic across underlying connections such as dedicated Internet, broadband, private lines, MPLS and wireless access. Learn more: Building a WAN →
WAN (Wide Area Network)
A network connecting users, applications and infrastructure across separate buildings, cities, countries or cloud environments. Learn more: Building a WAN →
Wavelength service
A carrier-managed, dedicated optical channel between two points. It provides high-capacity transport without requiring the customer to own the underlying fiber or operate the carrier’s optical platform. Learn more: Ethernet vs. Wavelength vs. Dark Fiber →
Data centers & colocation
AWS Direct Connect
Amazon Web Services’ private connectivity service for linking customer or colocation environments to AWS without relying solely on the public Internet. Learn more: Cloud Connectivity →
Azure ExpressRoute
Microsoft Azure’s private connectivity service for linking enterprise or colocation environments to Azure services through supported connectivity providers. Learn more: Cloud Connectivity →
Carrier hotel
A building where many telecommunications networks meet and interconnect, making it a key hub for routing traffic between carriers, clouds and enterprises. Examples include 350 E. Cermak in Chicago and 60 Hudson Street in New York.
Related: Meet-me room, interconnection, cross connect
COL2 (Cologix Columbus)
A Cologix colocation and interconnection facility in Columbus, Ohio. Firms can evaluate carrier and cloud access there against their own confirmed endpoints and network requirements. Related: colocation, interconnection.
Colocation
Housing customer-owned servers and network equipment in a third-party data center that supplies space, power, cooling, physical security and access to connectivity providers and interconnection ecosystems. Learn more: Data-Center Interconnection →
Data-center interconnection (DCI)
High-capacity connectivity between two or more data centers, colocation sites or critical facilities. DCI commonly supports replication, disaster recovery, workload mobility, storage and active-active applications. Learn more: Designing links between data centers →
DC2 (Equinix Ashburn)
An Equinix colocation and interconnection facility at 21715 Filigree Court in Ashburn, Virginia. It is a route-planning option for digital-asset infrastructure; a venue’s endpoint or matching-engine location must be confirmed separately. Learn more: Digital-Asset Connectivity →
Interconnection
The general practice of connecting networks, systems, providers or facilities. It can range from a cross connect within one building to a wavelength or dark-fiber route between cities. Learn more: Data-Center Interconnection →
Meet-me room (MMR)
The area inside a data center or carrier hotel where different networks physically connect to one another, usually by cross connects.
Financial markets
Digital-asset venue
A cryptocurrency exchange or trading platform where assets such as Bitcoin (BTC) and Ethereum (ETH) can be traded. Connectivity is increasingly evaluated using latency, capacity, resiliency and operational criteria similar to traditional venues. Learn more: Digital-asset and venue connectivity →
Direct feed vs. consolidated feed (SIP)
A direct feed is market data delivered straight from an exchange. A consolidated feed, produced by a securities information processor (SIP), combines quotes and trades from all U.S. exchanges into one stream. Direct feeds are faster; consolidated feeds are simpler and broader.
Related: Market data, market-data feed, market data symbol
Exchange or venue connectivity
Network infrastructure connecting a trading firm to a specific exchange or execution venue, typically through a cross connect or dedicated circuit into the venue’s data-center ecosystem. The New Jersey Equity Triangle is a major cluster of U.S. equities venues and related infrastructure. Learn more: Low-latency trading connectivity →
FR2 (Equinix Frankfurt)
An Equinix data center in Frankfurt, Germany, and the co-location site for Deutsche Börse's trading systems, including Eurex derivatives and Xetra cash equities on the T7 platform. FR2 is a central hub for European market-data distribution and interconnection, with routes commonly linking it to London (LD4) and New Jersey.
Related: LD4, matching engine, colocation, market-data feed
Free-space optics (FSO)
Laser-based links that transmit data through open air between two points in line of sight. FSO is used for short, very low-latency hops, often alongside millimeter wave, and can be affected by fog.
Related: Millimeter wave, microwave connectivity
Hollow-core fiber
Optical fiber that carries light through an air-filled core instead of solid glass. Because light travels faster through air, hollow-core fiber can reduce latency by up to about 30% compared with standard fiber over the same path.
Related: Dark fiber, latency, ultra-low-latency (ULL) connectivity
IT3 (Aruba Bergamo)
The Aruba Global Cloud Data Center in Bergamo, Italy, home of Euronext's core data centre since the exchange moved it from the UK in 2022. IT3 hosts matching engines and co-location for Euronext's markets, including Paris, Amsterdam, Brussels, Lisbon, Dublin, Oslo and Milan.
Related: Matching engine, colocation, LD4, FR2
LD4
An Equinix data center in Slough, near London, and a central hub for European trading, foreign-exchange and market-data connectivity. Transatlantic routes commonly link LD4 with the New Jersey trading sites.
Related: Equinix NY4 / NY5 / NY6, colocation, ultra-low-latency (ULL) connectivity
Market data
Pricing, order, trade and reference information distributed by exchanges, venues and data providers for use by trading, risk and analytics systems. Learn more: Market-Data Connectivity →
Market data symbol
An identifier for a specific tradable instrument within a market-data feed. Symbol count, message rates, feed depth and peak volume all influence the bandwidth and processing needed to receive data without gaps. Learn more: Financial Markets infrastructure →
Market-data feed
A defined stream or product carrying market information from an exchange, venue or distributor. Feed protocol, depth, symbols and peak message rate affect connectivity requirements. Learn more: Market-Data Connectivity →
Matching engine
The exchange system that pairs buy and sell orders and executes trades. Its physical location is the reference point for trading latency, which is why firms colocate as close to it as possible — for example, CME Globex in Aurora, Illinois, or NYSE in Mahwah, New Jersey.
Related: Colocation, CME Globex and GLink, NJ equities triangle, tick-to-trade latency
Microwave connectivity
A terrestrial wireless network that uses microwave radio links between towers or rooftops. On suitable routes it can reduce path distance compared with fiber, subject to geography, spectrum and operating conditions. Learn more: Low-Latency Trading Networks →
Millimeter wave (MMW)
High-frequency radio links that carry data through the air between towers or rooftops. Millimeter wave offers very low latency and more capacity than traditional microwave over short distances, but performance can degrade in heavy rain.
Related: Microwave connectivity, free-space optics, wireless backhaul
NJ equities triangle
The three northern New Jersey sites that host most U.S. equity matching engines: NYSE in Mahwah, Nasdaq in Carteret, and the Equinix campus in Secaucus. Firms trading across venues need fast, diverse connectivity between all three.
Related: Equinix NY4 / NY5 / NY6, matching engine, millimeter wave
Tick-to-trade latency
The time between receiving a market-data update and sending an order in response. It measures the full trading loop — network, data handling and trading systems — rather than network latency alone.
Related: Latency, market-data feed, matching engine
Ultra-low-latency (ULL) connectivity
Network infrastructure engineered to minimize and stabilize the time required to move market data and trade orders between precise endpoints, such as a trading firm, exchange, market-data source or execution venue. Learn more: Evaluating ultra-low-latency connectivity →
Wireless backhaul
A wireless connection carrying traffic between network sites, facilities or access points. Capacity, line of sight, spectrum, weather exposure and redundancy should be evaluated for the application. Learn more: Low-Latency Trading Networks →
Cloud & AI infrastructure
AI harness (also called agent harness)
The software layer that surrounds an AI model and turns it into a working system. A harness manages what the model can see and do: the tools and data it can access, memory across steps, permissions and guardrails, and how tasks are orchestrated and logged. The same model can behave very differently depending on the harness around it. For infrastructure planning, the harness determines where data flows, which systems the AI connects to, and what network, security and governance controls are needed — especially for private or on-premises AI deployments.
Related: Harness engineering, large language model (LLM), private AI environment, sovereign AI.
AI inference
The process of using a trained AI model to generate a prediction, classification, response or other output from new input data. Learn more: AI & Data Centers →
AI infrastructure
The compute, storage, networking, power, cooling, security and operating systems required to train, fine-tune or run AI workloads. Learn more: AI Data-Center Infrastructure →
AI training
The process of building an AI model by running it through large datasets on clusters of GPUs. Training is the most compute- and power-intensive AI workload and usually requires large, tightly connected GPU clusters.
Related: AI inference, GPU cluster, InfiniBand / RoCE
GPU (Graphics Processing Unit)
A processor designed for highly parallel computation and widely used for AI training and inference workloads. Learn more: AI Data-Center Infrastructure →
GPU cluster
A coordinated group of GPU-equipped servers connected by high-bandwidth, low-latency networking and supported by shared storage, power and cooling. Learn more: AI Data-Center Infrastructure →
Hybrid infrastructure
An environment combining on-premises systems, colocation facilities, private infrastructure and public-cloud resources through coordinated networking, security and operations. Learn more: Global and hybrid networking →
InfiniBand / RoCE
High-speed, low-latency networking technologies used to connect the GPUs in an AI cluster so they work as one system. InfiniBand is a specialized network; RoCE (RDMA over Converged Ethernet) delivers similar capabilities over Ethernet.
Related: GPU cluster, AI training, latency
Large language model (LLM)
An AI model trained on large amounts of text to understand and generate language. LLMs power chatbots, coding assistants and AI agents, and running them at scale drives demand for GPUs, power and high-capacity networks.
Related: Tokenization, AI inference, AI training, GPU
Private AI environment
AI compute, storage and networking operated within an organization’s controlled environment—on premises, in colocation or through dedicated hosted infrastructure—rather than relying exclusively on shared public-cloud resources. Learn more: Private AI Infrastructure →
Sovereign AI
AI infrastructure, data and models kept under the control of a specific country, company or organization, typically to meet data-residency, security or regulatory requirements.
Related: Private AI environment, AI infrastructure
Tokenization
A term with two common meanings in technology and finance.
In AI: the process of breaking text into tokens — words, parts of words or characters — that a model reads and generates. Tokens are the basic unit of AI workloads: models have token limits, providers price usage per token, and throughput is measured in tokens per second, which drives GPU, power and network requirements at scale.
In financial markets: representing ownership of an asset, such as a stock, bond, fund or real-world asset, as a digital token on a blockchain. Tokenized assets can settle faster and trade outside traditional market hours, creating new connectivity needs between digital-asset venues, custodians and traditional financial infrastructure.
Related: GPU, AI inference, large language model (LLM), digital-asset venue
Frequently asked questions
What is the difference between a cross connect and a private line?
A cross connect is a short physical cable connection between two parties within the same facility. A private line is a dedicated service that typically spans a longer distance between separate locations.
Is a cloud on-ramp the same as a cross connect?
Not exactly. A cloud on-ramp is a private access point into a cloud provider’s network. A cross connect is the physical in-building cable that may connect customer equipment to that on-ramp, a carrier, an exchange or another tenant.
Do I need dark fiber, or is a wavelength service enough?
A wavelength is often appropriate for a fixed amount of dedicated high-capacity transport. Dark fiber becomes more compelling when long-term growth, multiple wavelengths or direct control of the optical layer justifies the added equipment and operating responsibility.
What does colocation include besides space for servers?
Colocation facilities typically provide power, cooling, physical security, remote-hands options and access to carriers, other tenants, exchanges and cloud on-ramps through cross connects.
Related services
Related articles
Share the locations, applications, performance requirements and business objective. SpartaLink can help translate the terminology into practical options.
Discuss the Requirement