Glossary
Cloud, network and AI terms in plain English
Short definitions of the words you will meet in proposals, contracts and design documents.
A
- Anomaly detection
- Finding behavior that differs from a learned normal pattern, such as a sudden rise in latency on one path.
B
- Bandwidth
- The maximum rate a connection can carry data, usually measured in megabits or gigabits per second.
- BGP
- Border Gateway Protocol, the routing protocol networks use to exchange paths with each other, including cloud providers.
- Blueprint
- In VergForge, a saved, reusable combination of cloud and network building blocks with parameters teams can set.
C
- Cloud on-ramp
- A private entry point into a cloud provider’s network from your own network or a data center.
- Colocation
- Renting space, power and cooling for your equipment in a third-party data center.
- Cross-connect
- A physical cable inside a data center that links your equipment to a carrier or cloud provider.
D
- Data egress
- Data leaving a cloud provider’s network. Providers usually charge for it.
- Disaster recovery
- Plans and systems that restore operations after an outage, often by failing over to another site or region.
E
- Encryption in transit
- Protecting data while it moves between systems so it cannot be read if intercepted.
G
- GPU cloud
- A cloud service that rents graphics processors for AI training and inference.
H
- Hybrid cloud
- Running workloads across on-premises infrastructure and one or more public clouds.
I
- Inference
- Using a trained AI model to produce answers or predictions for new input.
- Infrastructure as code
- Defining infrastructure in version-controlled files so it can be created and changed automatically.
J
- Jitter
- Variation in delay between packets. High jitter harms voice and video.
L
- Latency
- The time data takes to travel from one point to another, usually measured in milliseconds.
M
- Micro-segmentation
- Dividing a network into small isolated zones with rules for what may talk to what.
- MPLS
- Multiprotocol Label Switching, an older private WAN technology many companies are replacing.
- Multi-cloud
- Using services from more than one public cloud provider.
P
- Packet loss
- Data packets that never reach their destination, which slows or breaks applications.
- Private interconnect
- A dedicated connection between your network and a cloud provider that does not use the public internet.
S
- SASE
- Secure Access Service Edge, a model that combines networking and security services delivered from the cloud.
- SD-WAN
- Software-defined wide area network, which steers traffic across multiple links using central policy.
- Segmentation
- Separating network traffic into groups, such as production and development, so they cannot reach each other freely.
- Service level agreement
- A contract that defines the performance and availability a provider commits to.
T
- Throughput
- The amount of data actually delivered over a connection in a given time.
- Training
- Teaching an AI model from data, which often requires moving large datasets to GPUs.
V
- Virtual circuit
- A logical private connection over shared physical infrastructure, kept separate from other traffic.
Z
- Zero trust
- A security approach that verifies every user and service on every request instead of trusting the network location.