Alibaba ACA Cloud Business key terms

Alibaba Cloud's core virtual machine service, letting you launch configurable compute instances with a chosen image, instance family, disk type and network setup, billed on subscription or pay-as-you-go terms.

A named ECS hardware profile such as general purpose, compute optimized or memory optimized, which fixes the ratio of vCPU to memory and suits particular workload shapes like web serving or in-memory caching.

A logically isolated private network you define inside Alibaba Cloud, with your own IP address range, subnets, route tables and gateways, giving you control over traffic separation between workloads.

A subnet created inside a VPC, tied to a single availability zone, from which cloud resources draw private IP addresses. Resources in different vSwitches can communicate when routing allows.

A stateful virtual firewall attached to ECS instances that permits or denies inbound and outbound traffic by rule, port and source. It acts as the primary instance-level network access control.

A public IPv4 address that can be allocated independently and bound to or unbound from cloud resources such as ECS instances or NAT gateways, enabling failover without changing the address customers use.

Alibaba Cloud's massively scalable object store for unstructured data such as images, backups and static website assets, organized into buckets and accessed over HTTP APIs with multiple storage classes.

The top-level container in OSS that holds objects. It has a globally unique name, a region, an access control policy and optional features such as versioning and lifecycle rules.

A tier of OSS or cloud disk storage balancing cost against retrieval latency and durability, ranging from frequently accessed standard tiers to infrequent, archive and cold archive tiers.

A policy on a storage bucket that automatically transitions objects to cheaper storage classes or deletes them after a set age, reducing cost for data whose access patterns decline over time.

A distributed edge caching service that serves static and media content from locations near end users, reducing latency and origin load by caching responses on points of presence worldwide.

A managed traffic distribution service that spreads incoming requests across backend servers using health checks and configurable algorithms, improving availability and horizontal scalability of applications.

A service that automatically adds or removes ECS instances based on schedules or metric thresholds, keeping capacity matched to demand and replacing unhealthy instances in a scaling group.

A fully managed relational database service supporting engines such as MySQL, PostgreSQL, SQL Server and MariaDB, handling provisioning, backups, patching and high-availability failover for you.

A replica of an RDS primary database that serves read queries, offloading reporting and read-heavy traffic from the primary node and improving overall query throughput.

A physically separate datacenter location with independent power and networking inside a region. Spreading resources across zones protects an application from a single-site failure.

A geographic area containing one or more availability zones where Alibaba Cloud resources are deployed. Choosing a region affects latency to users and data residency compliance obligations.

An architecture that runs workloads in two or more geographic regions, typically for latency reduction, regulatory data locality or resilience against a full regional outage.

The set of plans and technologies that restore service after a major failure, characterized by recovery point and recovery time objectives and by strategies such as backup-restore, pilot light or active-active.

The practice and tooling of copying data and system state to a separate location so it can be restored after loss or corruption, measured by how much data and time can be lost.

A design quality in which redundant components and automatic failover keep a system serving requests despite individual node, zone or network failures, usually expressed as an uptime target.

An architectural approach that separates application components so they interact through asynchronous interfaces such as queues or events rather than direct calls, improving resilience and independent scaling.

A style of building an application as small independently deployable services with clear boundaries, each owning its data and communicating over APIs or messaging instead of one monolithic process.

An application pattern where servers keep no session data between requests, so any instance can handle any request and instances can be freely added, removed or replaced.

A managed asynchronous messaging service that buffers messages between producers and consumers, smoothing traffic spikes, decoupling services and enabling reliable retry of failed processing.

A managed Kubernetes offering in which Alibaba Cloud operates the control plane while you run containerized workloads on node pools, integrating with networking, storage and scaling services.

A computing model where the provider manages servers and scales capacity automatically, and you pay only for resources consumed during execution rather than for idle provisioned machines.

An event-driven serverless compute service where you deploy code that runs on demand in response to triggers such as HTTP requests, object uploads or scheduled events, billed per invocation and duration.

A managed network component that lets instances in private subnets reach the internet for outbound traffic without exposing them to inbound connections, and can also forward inbound traffic.

A dedicated private network connection between an on-premises datacenter and Alibaba Cloud, offering more consistent latency and bandwidth than internet-based links for hybrid workloads.

An environment combining on-premises infrastructure with public cloud resources, connected by private or encrypted links, so workloads can span both while sharing identity, networking and data.

The structured process of moving applications, data and dependencies from existing infrastructure into the cloud, typically involving assessment, planning, transfer, validation and cutover phases.

A tool that replicates data from a source database to a target with schema conversion and continuous synchronization, allowing a low-downtime cutover during a move to managed databases.

The full lifetime cost of running a workload, including compute, storage, network egress, licensing, operations labor and downtime, used to compare cloud against on-premises options fairly.

A billing model in which resources are charged by actual usage over time with no long commitment, giving flexibility for variable or short-lived workloads at a higher unit price.

A billing model where capacity is reserved for a fixed term, often a year or more, in exchange for a lower unit price than on-demand rates, suited to steady predictable workloads.

Alibaba Cloud's identity and permission service for creating users, groups and roles and granting fine-grained policies over resources, including temporary credentials for applications.

A neural network trained on vast text corpora to predict and generate language, capable of tasks such as summarization, translation, question answering and code generation from natural-language prompts.

Alibaba Cloud's platform for accessing and building with foundation models, offering hosted model APIs, fine-tuning, prompt tooling and application orchestration for generative AI solutions.

A technique that retrieves relevant documents from a knowledge base and supplies them to a language model with the question, grounding answers in private or current data and reducing fabrication.

A numeric vector representation of text, images or other content in which semantically similar items sit close together, enabling similarity comparison and search over unstructured data.

A retrieval method that finds items by comparing embedding vectors for similarity rather than matching keywords, powering semantic search and knowledge retrieval for generative AI applications.

Further training of a pretrained foundation model on a smaller domain-specific dataset so its behavior and outputs better match a particular task, tone or proprietary knowledge.

The practice of designing instructions, examples and context in a model input to steer output quality, format and reliability without changing the model's weights.

A failure mode where a generative model produces fluent but unsupported or incorrect content, which architects mitigate with grounding, retrieval, validation and human review.

The unit into which text is split for language model processing and billing. Cost and context limits are usually expressed in tokens rather than characters or words.

A configurable control around a generative AI application that filters unsafe or off-policy inputs and outputs, enforces content rules and limits topics the model may respond to.