Dell DCA-ISM flashcards

31 cards. Read the term and the definition here. The deck at the top of this page is optional.

Data vs. information

Data is raw unprocessed facts; information is data processed and placed in context so it becomes meaningful for decisions. Storage persists both, but only processed data supports business action.

Structured vs. unstructured data

Structured data fits a fixed schema such as relational rows and columns. Unstructured data has no fixed schema, including documents, images, and logs. Semi-structured data like JSON sits between the two.

Block, file, object access

Block access addresses raw fixed-size blocks through LUNs. File access addresses named files through a filesystem. Object access addresses whole objects by identifier and metadata over HTTP APIs.

Write-back versus write-through cache

Write-back cache acknowledges the host after the write reaches cache, which is fast but requires mirroring or power protection. Write-through acknowledges only after the data reaches disk, which is safer but slower.

RAID 5 usable capacity

RAID 5 usable capacity equals (number of drives minus one) multiplied by drive size, because one drive's worth of capacity is consumed by distributed parity. It tolerates one drive failure.

RAID 10 characteristics

RAID 10 mirrors stripes of data. It delivers strong random-write performance and fast rebuilds, tolerates one failure per mirrored pair, and halves usable capacity relative to raw drives.

Fibre Channel zoning

Zoning controls which FC ports or WWNs may communicate in a fabric. Hard zoning enforces by physical port; soft zoning enforces by WWN. Single-initiator, single-target zones simplify troubleshooting.

iSCSI initiator and target

In iSCSI the host runs an initiator that sends SCSI commands over TCP/IP to a storage target. CHAP authenticates the session, and MPIO provides redundant paths.

Object storage identifier

Each object consists of payload data, metadata, and a globally unique identifier. Objects live in a flat namespace, typically a bucket, and are retrieved by key over REST APIs.

Erasure coding overhead

Erasure coding splits data into k data fragments and m parity fragments. It tolerates m simultaneous failures, and storage overhead equals (k plus m) divided by k.

Object immutability and versioning

Most object stores treat objects as immutable, so updates create new versions. Versioning plus object lock supports retention and recovery from ransomware or accidental deletion.

Unified storage definition

Unified storage serves block, file, and often object protocols from one system with shared management and a common capacity pool, simplifying operations for mixed workloads.

Software-defined storage

SDS separates the control plane, which handles policy and provisioning, from the data plane, which handles I/O, running on commodity hardware with automation through APIs.

Hyperconverged infrastructure

HCI bundles compute, storage, and virtualization into one cluster of nodes. Storage is distributed across nodes and protected with replication or erasure coding rather than shared shelves.

RPO versus RTO

Recovery point objective is the maximum acceptable data loss measured in time. Recovery time objective is the maximum acceptable outage duration. Both drive the choice of replication and backup design.

Synchronous versus asynchronous replication

Synchronous replication acknowledges writes only after the remote site confirms, giving near-zero RPO but requiring low latency. Asynchronous replication acknowledges locally, tolerating distance at the cost of nonzero RPO.

Snapshot versus backup

A snapshot is a fast point-in-time metadata copy, often using copy-on-write. It protects against logical errors but resides on the same system, so it does not replace an offsite backup.

Encryption at rest

Encryption at rest protects stored data through array-level, host-level, or backup encryption. Key management is the critical dependency, because lost keys make the data unrecoverable.

LUN masking

LUN masking restricts which host initiators can see a given block LUN. It is an access control on the storage array, not encryption, and works together with fabric zoning.

Ransomware storage defenses

Effective defenses include immutable snapshots, object lock or WORM copies, an offline air-gapped backup, least-privilege accounts, and regularly tested restore procedures.

Thin provisioning risk

Thin provisioning presents logical capacity larger than physical capacity. Without monitoring and space reclamation, the pool can exhaust, causing write failures across all volumes sharing it.

Little's Law for storage

Queue depth equals IOPS multiplied by latency. This relationship helps validate reported performance metrics and identify whether a bottleneck is queuing or service time.

FC fabric health indicators

Key indicators include ISL utilization, buffer-to-buffer credit exhaustion, CRC errors, link resets, and zoning consistency across switches. Rising CRC errors usually indicate faulty optics or cabling.

Cloud service models

IaaS provides virtual machines and storage, PaaS provides managed runtimes and databases, and SaaS provides finished applications. Customer management responsibility decreases from IaaS to SaaS.

Cloud storage classes

Cloud object stores offer tiers such as hot, infrequent access, and archive. Lower-cost tiers add retrieval fees and minimum storage durations, so lifecycle rules should match access patterns.

Cloud shared responsibility

The provider secures physical infrastructure and the service platform. The customer secures data, identities, encryption keys, access policies, and configuration of the services used.

AI training versus inference I/O

Training is throughput-bound on large sequential reads of datasets and periodic checkpoint writes. Inference is latency-bound on small random reads, so the two stages need different storage tiers.

NVMe over Fabrics

NVMe-oF carries the NVMe command set over RDMA, TCP, or Fibre Channel fabrics. It provides remote flash latency far below legacy SCSI paths, enabling disaggregated storage.

Data gravity

Large datasets attract applications and services toward their location because moving data is costly. Placement decisions for analytics and AI should account for where the data already resides.

Container Storage Interface

CSI is a standard plugin interface that lets Kubernetes dynamically provision and attach persistent volumes from external arrays, software-defined storage, or cloud storage.

Composable infrastructure

Composable infrastructure pools compute, storage, and network resources and assigns them to workloads through software orchestration, improving utilization but adding operational complexity.