NCP-ADS key terms

A two-tier data center network design where every leaf switch connects to every spine switch, giving any server a predictable number of hops to any other server and enabling even traffic distribution across uplinks.

Network traffic that flows between servers inside the data center, such as storage replication or application-to-database calls, as opposed to traffic entering or leaving the site.

Equal-Cost Multi-Pathing: a routing technique that spreads flows across several equal-cost links simultaneously, increasing usable bandwidth and resilience in a fabric with redundant paths.

Bonding multiple physical network interfaces into one logical link so traffic shares the combined bandwidth and the connection survives the failure of an individual member port.

Ethernet frames with a payload larger than the standard 1500 bytes, commonly about 9000 bytes, which reduce per-packet overhead for storage and backup traffic when every device on the path agrees.

A logical Layer 2 broadcast domain carved out of shared physical switching hardware, letting administrators isolate traffic groups such as management, storage, and tenants without separate cabling.

A network virtualization encapsulation that carries Layer 2 frames inside UDP packets across a Layer 3 underlay, allowing large numbers of isolated overlay segments to span routed data center networks.

A virtual network built as tunnels on top of an existing physical underlay, so logical connectivity and segmentation can be defined in software independently of the cabling and routing beneath.

A centralized software component that programs forwarding behavior across network devices through APIs, replacing per-device manual configuration and enabling policy to follow workloads automatically.

A disk array layout that stores two independent parity blocks across member drives, so the array keeps running through the failure of any two disks while sacrificing two drives' worth of capacity.

A mirrored-then-striped layout that pairs disks and stripes across the pairs, delivering strong random-write performance and fast rebuilds at the cost of losing half the raw capacity.

A standby drive kept powered and connected but unused, which the storage system automatically brings into a degraded array to begin rebuilding without waiting for a technician.

The period during which a storage array reconstructs data onto a replacement drive; a longer window raises the risk that another disk fails before redundancy is restored.

Allocating logical capacity to volumes on demand rather than reserving physical space up front, which improves utilization but requires monitoring so the underlying pool never runs out.

Logical Unit Number: an addressable block storage volume presented from a storage system to a host, which the host formats with a file system or uses directly for applications.

A protocol that transports SCSI block storage commands over ordinary IP networks, letting hosts attach remote volumes without dedicated fibre channel hardware or a separate storage fabric.

An extension of the NVMe command set across a network transport such as RDMA or TCP, giving remote storage latency much closer to locally attached flash than older storage protocols.

Placing data automatically on media of different cost and speed, such as flash for hot blocks and spinning disk for cold blocks, based on observed access patterns over time.

Storing data as fragments plus computed parity chunks across many drives or nodes so the original survives several losses, typically with less capacity overhead than full mirroring.

Continuously copying data from one storage system or site to another so a usable duplicate exists; synchronous replication waits for the copy, while asynchronous replication does not.

A point-in-time logical image of a volume that preserves the state of its blocks, used for quick rollback or as the source for backups without copying the entire dataset immediately.

Recovery Point Objective: the maximum acceptable amount of data, measured in time, that a recovery may lose, which drives how frequently backups or replication must run.

Recovery Time Objective: the target elapsed time within which a service must be restored after an outage, which determines the recovery architecture and automation required.

A backup copy that cannot be altered or deleted for a defined retention period, protecting recovery data from ransomware encryption, credential theft, and accidental deletion.

A deliberate physical or logical separation preventing direct network access to a copy of data, so an attacker who compromises production systems cannot reach the protected copy.

The process of shifting service from a failed or degraded component or site to a standby one, ideally automated so that recovery completes within the defined objective.

A cluster condition where two nodes each believe they are authoritative, often after a network partition, risking divergent writes and corruption unless quorum or fencing resolves it.

The software layer that runs virtual machines, allocating CPU, memory, storage, and network resources to guests while isolating them from each other and from the underlying hardware.

Moving a running virtual machine between hosts with little or no downtime by transferring its memory state while the workload continues to execute.

A placement policy that keeps specified virtual machines together on the same host, or apart on different hosts, to meet licensing, performance, or availability requirements.

The uncontrolled growth of virtual machines, including forgotten or orphaned instances, which consumes storage and compute and complicates patching, licensing, and security review.

Managing servers, networks, and platform settings through version-controlled definition files that tools apply consistently, instead of configuring systems by hand through consoles.

A property of an operation that produces the same end state whether run once or many times, allowing automation to be re-applied safely without duplicating or corrupting resources.

Describing the desired end state of a system rather than the step-by-step commands to reach it, letting the tool decide what changes are actually needed.

A practice in which servers are replaced with newly built images for every change rather than modified in place, reducing configuration drift and making rollback predictable.

The gradual divergence between a system's actual settings and its intended or documented state, often caused by manual changes, which undermines reproducibility and troubleshooting.

A documented reference measurement of normal system behavior, such as typical latency or utilization, used to judge whether later observations represent a genuine problem.

A latency statistic such as p95 or p99 showing the value below which that share of requests complete, which reveals tail behavior that averages hide.

The number of I/O requests outstanding at a storage device or array; sustained high values indicate that demand exceeds service capacity and latency will climb.

The desensitization that occurs when monitoring generates excessive or low-value alerts, causing responders to overlook or delay action on a genuine incident.

Service Level Objective: a measurable internal target for a service, such as availability or response time, used to judge whether reliability work is needed and to prioritize it.