NCP-ADS flashcards
NCP-ADS · Build recall with a focused review session.
32 cards. Read the term and the definition here. The deck at the top of this page is optional.
- Oversubscription ratio
- The ratio of total downstream bandwidth to total upstream bandwidth in a network tier. A 3:1 ratio means connected hosts can collectively demand three times the available uplink capacity.
- Leaf-spine topology
- A two-tier data center network where every leaf switch connects to every spine switch. Any inter-leaf path is exactly two hops, giving predictable latency and equal-cost multipathing.
- VXLAN VNI
- A 24-bit VXLAN Network Identifier carried in the encapsulation header. It supports roughly 16 million isolated Layer 2 segments, far exceeding the 4094 limit of 802.1Q VLAN IDs.
- Jumbo frames
- Ethernet frames with an MTU of 9000 bytes instead of the standard 1500. Every device on the path must support the same MTU, or large packets are silently dropped.
- East-west traffic
- Network traffic between servers inside the data center, such as application-to-database calls. It typically dominates volume and drives spine capacity planning.
- RAID 6 usable capacity
- Usable capacity equals (number of disks minus two) multiplied by disk size. Two parity disks allow the array to survive two simultaneous drive failures.
- Write amplification
- The ratio of bytes written to flash memory divided by bytes written by the host. Garbage collection and partial-page writes inflate it, reducing SSD endurance and steady-state performance.
- Thin provisioning
- Presenting more logical capacity to hosts than physically exists. It improves utilization but requires monitoring of the overcommit ratio to prevent out-of-space events.
- IOPS versus throughput
- IOPS counts operations per second and governs small random workloads. Throughput measures bytes per second and governs large sequential workloads such as backup and streaming.
- NFSv4 statefulness
- NFSv4 keeps connection state, uses a single TCP port 2049, and integrates locking and ACLs. NFSv3 is stateless and relies on separate auxiliary protocols for locking.
- iSCSI target port
- iSCSI carries SCSI commands over TCP, conventionally on port 3260. Initiators connect to targets to access block LUNs across standard IP networks.
- Recovery point objective
- RPO is the maximum acceptable amount of data loss measured in time. A 15-minute RPO means backups or replication must occur at least every 15 minutes.
- Recovery time objective
- RTO is the maximum acceptable elapsed time to restore a service after an outage. It drives choices such as hot, warm, or cold recovery sites.
- 3-2-1 backup rule
- Keep three copies of data, on two different media types, with one copy offsite. An immutable or offline copy additionally defends against ransomware encryption.
- Differential backup
- A backup that copies all data changed since the last full backup. Restoring requires only the last full plus the latest differential, keeping the restore chain short.
- Incremental backup
- A backup that copies only data changed since the most recent backup of any type. Restores require the full plus every incremental in sequence, lengthening recovery time.
- Synthetic full backup
- A full backup image assembled from the previous full plus subsequent incrementals, without re-reading production data. It shortens backup windows while keeping restore chains short.
- Application-consistent snapshot
- A snapshot taken after quiescing the application, often via VSS or hypervisor tools, so in-flight writes are flushed. Crash-consistent snapshots may require recovery or log replay.
- Type 1 hypervisor
- A hypervisor that runs directly on server hardware, such as ESXi or Hyper-V. It avoids a host operating system layer and typically delivers lower overhead than Type 2.
- Memory ballooning
- A hypervisor technique that reclaims guest memory by inflating a driver inside the guest, forcing it to release pages. It is preferred over swapping to hypervisor disk.
- Live migration prerequisites
- Live migration requires shared storage or storage motion, compatible CPU features across hosts, and sufficient network bandwidth. Incompatible CPUs block migration or force a compatibility mask.
- Shared responsibility model
- In cloud computing, the provider secures physical facilities, hardware, and the hypervisor. The customer secures guest operating systems, identities, data, and configuration.
- Availability zone
- An isolated failure domain within a cloud region, with independent power and networking. Deploying across zones removes single-datacenter failure as a cause of outage.
- Declarative infrastructure as code
- A model where you describe the desired end state and the tool converges actual infrastructure to match. It is idempotent, unlike imperative scripts that list explicit steps.
- Idempotency
- The property that running the same automation repeatedly produces the same result without creating duplicates. It is the core requirement for safe, repeatable configuration management.
- Terraform state locking
- A mechanism that prevents concurrent applies from modifying the same state file. Remote backends with locking avoid corruption and conflicting resource changes.
- Configuration drift
- A divergence between the declared or expected configuration and the actual state of infrastructure. Detection compares live resources against state or a compliance baseline.
- Immutable infrastructure
- A practice of replacing servers rather than modifying them in place. New instances are built from versioned images, which eliminates configuration drift and simplifies rollback.
- USE method
- A performance methodology that checks Utilization, Saturation, and Errors for every resource. It quickly identifies which subsystem is constrained before deeper investigation.
- Four golden signals
- Latency, traffic, errors, and saturation. Monitoring these four signals for each service gives broad coverage of user-visible health with relatively few metrics.
- Little's Law
- Concurrency equals arrival rate multiplied by latency. At fixed throughput, rising latency indicates growing queue depth and approaching saturation.
- Tail latency
- Latency measured at high percentiles such as p95 or p99. It exposes the slowest requests that averages hide and that users actually experience.