Cloud Storage Glossary

Mechanisms that define who can access data and what actions they can perform, ensuring data security and compliance.

Fine-grained permissioning using Identity and Access Management. Example: AWS IAM policies for bucket-level access.

Storage tuned for workload patterns: sequential reads, random access, big data analytics, or AI training.

High-throughput, low-latency storage designed for training/inference pipelines with GPUs or HPC clusters.

Application Programming Interfaces that allow developers to interact programmatically with cloud storage services.

Data is first written to primary location, then propagated to replicas. Reduces latency but introduces a small risk of data loss if primary fails.

Tracks access and modifications for regulatory adherence. Example: S3 access logs, CloudTrail, Azure Monitor.

The uptime and accessibility of data, expressed as a percentage. SLAs differ by storage class; e.g., S3 Standard offers 99.99% availability.

Time to attach volumes to nodes and mount behavior (lazy vs immediate mounts); impacts pod startup time.

Cloud-based service that provides backup solutions, allowing businesses to back up their data to the cloud.

Data storage where data is stored in fixed-size blocks, each with its own address, commonly used for databases and virtual machines.

Memory-efficient membership tests to reduce disk lookups for absent keys (used in object stores / indexes).

Automatic redistribution of data when nodes join/leave and self-healing after corruption or disk failures.

Standard tools to measure IOPS, throughput, and latency under controlled workloads.

Tools/operators for orchestrated backups, restores, and policy-driven protection in Kubernetes environments.

Validates stored data using checksums or hash algorithms to detect corruption.

A model of data storage where digital data is stored in logical pools across multiple servers, often in different locations, managed by a hosting company.

Integration with compute, analytics, AI/ML, and serverless services. Example: S3 + EMR or GCS + Vertex AI.

Integration with compute, analytics, and AI/ML services. Example: AWS S3 + EMR for big data processing, GCP Storage + Vertex AI for ML training.

A storage tier optimized for infrequently accessed data, offering lower costs but higher retrieval times, often used for archival purposes.

A storage class designed for data that is rarely accessed, offering low-cost storage with higher retrieval times, suitable for archival data.

Adherence to regulatory requirements and standards (e.g., GDPR, HIPAA) governing data storage and protection.

Reducing data size to save storage space and bandwidth. Applied in block or object storage for efficiency.

Rules governing how and when data updates are visible across distributed systems. Types: strong consistency, eventual consistency, read-after-write consistency.

Cross-zone: within same region (low latency), Cross-region: across regions for DR and geo-redundancy.

Algorithms that ensure distributed metadata/state agreement (used by etcd, Ceph MONs) for leader election and consistent cluster state.

How large objects are split into chunks/segments—affects parallelism, repair, and I/O performance.

Background process that rewrites data structures to reduce fragmentation and reclaim space (common with LSM).

Snapshot/clone implementation patterns that avoid full copies by sharing/redirecting blocks on write.

Controller (control-plane ops like create/delete) and node (attach/mount) side components implementing CSI spec.

Local caches (LRU, TTL-based) for hot data/metadata to reduce backend IOPS and lower latency.

Choice impacts CPU cost and collision risk; CRC32C common for perf, SHA256 for cryptographic integrity.

Inject faults (disk loss, network partition) to validate healing, redundancy, and SRE runbooks.

Indexing/search and audit stores for governance, discovery, and compliance across the object namespace.

The process of moving data that is no longer actively used to a separate storage device for long-term retention.

Reducing the size of data to save storage space and improve transfer speeds, often used in cloud storage solutions.

Eliminating duplicate copies of repeating data to save storage space and improve efficiency.

Ensuring that data is accurate, consistent, and unaltered during storage and transmission.

The process of transferring data between storage systems or locations, often during cloud adoption or infrastructure upgrades.

Transferring data to the cloud or between storage systems. Example: AWS DataSync, Google Transfer Service.

The practice of storing copies of data across multiple locations to ensure availability and durability in case of hardware failure.

Periodic process to verify and correct data integrity across storage nodes.

The concept that data is subject to the laws and regulations of the country in which it is stored, impacting cloud storage decisions.

The practice of moving data between different storage types based on access patterns and cost considerations.

Charges for moving data out of cloud regions; critical for global enterprise workloads.

Removing duplicate data to save storage space and reduce costs. Common in backup and archival systems.

Strategies and tools to recover data and applications in the event of a catastrophe, ensuring business continuity.

Strategies to switch operations to backup systems during failure (failover) and revert after recovery (failback).

Storage system that spreads data across multiple nodes for horizontal scalability and high availability. Used in cloud-native and big data workloads. Example: Ceph, Amazon FSx.

Probability that data remains intact and uncorrupted over time. Enterprise-grade storage targets 11–16 nines of durability. Example: S3 Standard offers 99.999999999% durability.

Periodic verification of checksums to detect silent corruption and trigger repairs.

Columnar / table formats enabling efficient analytics on object storage with partitioning and ACID semantics.

Scheduling pods or jobs near data replicas to reduce network hops and maximize throughput.

Storing data closer to the location where it is needed to reduce latency and bandwidth usage, often used in IoT applications.

Caches data closer to users/devices for low-latency access. Used in IoT, streaming, and AI inference.

The process of converting data into a coded format to prevent unauthorized access, both at rest and in transit.

Protecting data using cryptographic methods. At-rest uses AES256/KMS keys; in-transit uses TLS/SSL. Required for compliance standards (HIPAA, PCI, GDPR).

Advanced redundancy technique splitting data into fragments with parity across nodes. Reduces storage overhead while maintaining durability.

Trigger actions when objects are created, updated, or deleted. Example: S3 events triggering Lambda.

Redundancy scheme that slices data into k data + m parity shards to reduce storage overhead vs replication while allowing reconstruction.

Weaker models where replicas converge over time; useful for high-availability geo-replication.

Reconstructing lost shards optimally using local parity to minimize cross-rack/regional traffic.

Hierarchical storage system using directories and files. Accessed via protocols like NFS or SMB. Ideal for shared enterprise applications or CI/CD pipelines.

Whether write calls are flushed to stable storage immediately—critical for databases (fsync cost vs durability).

Data is stored in multiple geographic regions to prevent data loss due to regional disasters. Critical for global SaaS and DR strategies. Example: AWS S3 Cross-Region Replication.

A storage tier designed for frequently accessed data, providing high performance and low latency, suitable for active applications.

Combines on-premises and cloud storage for latency-sensitive or regulatory-bound workloads. Example: Azure Arc-enabled storage.

Multi-media architectures where media class influences latency, throughput and tiering policies.

Prevents accidental or malicious changes. Used for compliance logs, blockchain storage, or critical archives.

Prevents modification/deletion after writing. Used for compliance, financial, or audit data.

On-the-fly compression reduces storage but adds CPU; choose algorithm as tradeoff between ratio and latency.

Removing duplicate blocks/objects (fingerprinting via hashes) to save capacity—requires metadata index.

Centralized management of encryption keys for enterprise-grade storage. Example: AWS KMS, Azure Key Vault.

The time delay between a user’s request and the data’s response, crucial for performance-sensitive applications.

Key performance metrics: Latency = response time; IOPS = read/write operations/sec; Throughput = data volume/sec.

Automating the movement of data between different storage classes based on predefined rules, optimizing cost and access speed.

Storage engine data structures used for indexing (LSM for write-heavy workloads; B-tree for balanced reads/writes).

Distributing data across multiple geographic locations to enhance availability and disaster recovery capabilities.

Dedicated service or sharded store for object/file metadata (namespaces, inodes, object indices) to scale lookups.

Breaking large uploads into parts for parallel transmission and resumability (S3 multipart semantics).

Architectures combining fast in-memory caches and persistent sharded index for scalable metadata ops.

A storage tier that balances cost and access speed, used for data that is accessed less frequently but still requires quick retrieval.

A protocol that allows file access over a network, enabling a system to share directories and files with others over a network.

High-performance transport stacks (NVMe over Fabrics, RDMA, SPDK) to reduce latency and CPU overhead for NVMe devices.

Low-level NVMe constructs used to slice devices for isolation or QoS.

OS/kernel-level and service-level controls to isolate noisy tenants and enforce fairness.

A storage architecture that manages data as objects, each containing the data itself, metadata, and a unique identifier, ideal for unstructured data like media files.

Prometheus exporters and metrics (ops/sec, avg latency, queue depth, rebuild backlog) for monitoring.

Policy primitives for retention, legal holds, and immutable retention (WORM) with enforcement at store level.

Billing model based on actual usage instead of provisioned capacity.

Measures storage performance: IOPS for block storage, throughput for bulk transfer, latency for real-time apps.

Persistent storage retains data after system shutdown (e.g., EBS, S3), ephemeral storage is temporary and deleted with the instance (e.g., EC2 instance store).

Logical groupings that control how objects/chunks are distributed across racks/hosts to avoid correlated failures.

POSIX (strong namespace, byte-level updates) vs object (immutable objects with PUT/GET) — choose based on app needs.

Placement group mechanics (e.g., Ceph PGs) for mapping objects to OSDs and how rebalancing is triggered.

Byte-addressable storage (e.g., Intel Optane) for ultra-low latency persistence or fast metadata stores.

Tail-latency metrics and objectives (percentile-based) used to drive SLOs and error budgets.

Minimum set of replicas required to accept reads/writes to guarantee consistency under failures.

Mechanisms to cap or reserve throughput and IOPS per tenant/volume to guarantee SLAs.

Storage compliance with HIPAA, PCI DSS, SOC2, GDPR, etc.

The process of copying data from one location to another to enhance data availability and fault tolerance.

An architectural style for designing networked applications, using HTTP requests to access and use data, commonly used in cloud storage services.

Maximum tolerable data loss measured in time. Defines how frequently backups or replication should occur. Example: 15-minute RPO means only 15 minutes of data can be lost.

Maximum tolerable downtime before systems must be restored. Example: 2-hour RTO means systems must be recoverable within 2 hours.

Number of full copies maintained for data; affects durability, read throughput, and storage overhead.

Guarantee that a write is immediately visible to subsequent reads—important for many transactional workloads.

Kubernetes PV reclaim behavior after PVC deletion—controls data lifecycle and accidental deletion protection.

Network and disk bandwidth available for reconstructing lost shards—affects recovery time (RTO).

Traffic shaping algorithm often used to enforce egress limits or burst control.

The ability of a storage service to support Amazon S3’s API, enabling interoperability with S3 tools and applications.

Formal guarantee on storage uptime, durability, and support response. Crucial for mission-critical enterprise apps.

A network file sharing protocol that allows applications to read and write to files and request services from server programs.

Read-only copies of storage volumes at a specific time. Used for backups, versioning, and disaster recovery.

Read-only copies of storage at a specific time. Used for backups, rollback, or DR.

A protocol for exchanging structured information in the implementation of web services, used in some cloud storage APIs.

A cloud computing model where a service provider rents out storage resources to customers on a subscription basis.

Data is written simultaneously to multiple locations before confirming success. Ensures zero data loss, used for mission-critical workloads.

Read-after-write guarantees where operations appear instantaneous and globally ordered.

Block-level snapshot and cloning APIs; implementation can be hardware-assisted or software-copy-on-write.

Techniques (inline metadata, packed objects) to reduce overhead for many small objects and improve performance.

Engineering-level objectives with documented remediation steps and playbooks for breaches.

Server-side querying of objects (e.g., Parquet/CSV) to reduce data egress and speed analytics.

The amount of data transferred over a network in a given time period, affecting the speed of data access and transfer.

Storage architecture that automatically moves data between hot, warm, and cold tiers based on access patterns and lifecycle policies. Optimizes cost and performance.

Marking deleted objects and later reclaiming storage; GC latency impacts storage footprint and consistency.

Thin: allocate on-demand (saves capacity); Thick: allocate up-front (predictable performance).

Distributed tracing of storage operations (client → metadata → data shards) to debug performance issues.

High-performance storage optimized for low latency and high IOPS, often backed by NVMe SSDs for AI/ML and HPC workloads.

The ability to keep multiple versions of an object or file, allowing recovery of previous states and protection against accidental deletions.

Kubernetes storage topology rules (Immediate vs WaitForFirstConsumer) and node/zone-aware provisioning for locality.

Storage tier for moderately accessed data. Balanced between cost and access speed. Example: weekly reports or infrequently queried logs.

Storage type that prevents modification/deletion after writing. Used for compliance or financial archives.

Extra physical I/O incurred for logical writes/reads due to metadata, replication, or compaction—key perf metric.

Write-back buffers writes and ack earlier (higher perf, more risk); write-through writes synchronously for safety.