Product overview

更新时间:
复制 MD 格式

Alibaba Cloud Container Compute Service (ACS) provides serverless, Kubernetes-compliant container compute resources without cluster or node O&M. ACS supports diverse containerized application and cloud service workloads.

Product introduction

What is Container Compute Service (ACS)?

Container Compute Service (ACS) is an upgrade of ACK Serverless clusters (FKA ASK), offering improved cost-efficiency, usability, and elasticity. ACS defines serverless compute classes and QoS classes, enabling on-demand resource requests billed per second without cluster or node O&M.

Comparison item

ACS cluster

ACK Pro cluster

Features

  • Control planes are created and hosted by ACS.

  • No need to design cluster size, choose node specifications, create nodes, or manage nodes.

  • Apply for pods on demand. ACS automatically allocates compute resources, eliminating node O&M.

  • Control planes are created and hosted by ACK.

  • You can flexibly design the cluster size and choose node specifications.

  • You must create nodes before deploying pods. You need to maintain nodes and optimize resource utilization.

Billing

No cluster management fee. Pods are billed per second based on instance type and allocated vCPU, memory, or GPU resources. Billing.

A cluster management fee is charged. Nodes are billed based on specifications and uptime. For more information, see Billing overview.

ACS compute classes

ACS defines compute classes for CPU and GPU compute power, simplifying resource allocation and cluster sizing. Overview of ACS pods.

Compute class: general-purpose

Compute class: performance-enhanced

Compute class: GPU-accelerated

GPU-HPN

QoS class: default

Supported

Supported

Supported

Supported

QoS class: best-effort

Supported

Supported

Supported

Not supported

Note

GPU compute power is in invitational preview. Contact your sales manager or PDSA for access.

Benefits

  • Cost-effectiveness and ease of use

    ACS provides general-purpose and performance-enhanced pods for online workloads, and cost-effective BestEffort pods for offline workloads. Deploy pods quickly using YAML templates or the ACS console.

  • Fine-grained and elastic resource requests

    CPU resources start at 0.25 vCPUs and 0.5 GiB memory, scaling in increments of 0.5 vCPUs and 1 GiB. GPU resources start at one GPU. Request only what you need to reduce costs.

  • On-demand scaling and pay-as-you-go billing

    Scale workloads within seconds and request elastic resources on demand. Pay-as-you-go billing is per second. Savings plans based on daily committed consumption further reduce costs for fluctuating workloads.

  • Simplicity and rich scenarios

    ACS hosts key Kubernetes system components and auto-updates cluster patch versions, reducing O&M complexity. Supports containerized applications and cloud service workloads across diverse scenarios.

Scenarios

  • Common online businesses

    For online applications such as microservices and web apps, use general-purpose pods. Launch and scale workloads within seconds to handle traffic spikes, preventing business loss without over-provisioning.

  • Big data computing businesses

    For latency-tolerant, throughput-intensive workloads such as Spark, Presto, and AI training jobs, use cost-effective BestEffort pods. Scale within seconds to improve parallel computing efficiency.

  • AI training and inference businesses

    For latency-sensitive AI workloads such as AIGC model training and inference, autonomous driving model training and inference, and real-time inference, combine GPU-accelerated pods with GPU or GPU-HPN capacity reservations to guarantee resource supply and reduce costs.

  • High-performance businesses

    For high-performance workloads such as cloud gaming, scale within seconds to handle traffic spikes, improve processing speed, reduce latency and stutters, and ensure optimal user experience.

Key features

Resource management

Feature

Description

Pod types

ACS provides general-purpose, performance-enhanced, GPU-accelerated, and GPU-HPN pods for different business scenarios. Overview of ACS pods.

On-demand scaling

ACS scales resources on demand by pod type, billed per second on a pay-as-you-go basis. View metering data on the billing details page.

Capacity reservations

ACS provides pod capacity reservations and node capacity reservations for pods that run GPU inference businesses.

Cluster management

Feature

Description

Cluster creation

ACS clusters use serverless resources, eliminating cluster and node maintenance. Create and deploy workloads using YAML templates or the ACS console.

Cluster connection

You can obtain the kubeconfig file of an ACS cluster and then use kubectl to connect to the ACS cluster to manage ACS resources.

Authorization management

ACS allows you to manage RAM permissions for resources and RBAC permissions for Kubernetes clusters. For more information, see Authorization overview.

Scheduling domain management

ACS is compatible with Kubernetes scheduling, supports colocation of multiple workload types, and provides fine-grained scheduling for elastic and heterogeneous resources.

Note

After you log on to the ACK console, you can view ACS clusters, manage RBAC permissions, and manage kubeconfig files.

Application management

Feature

Description

Application creation

ACS supports a variety of workloads, including Deployments, StatefulSets, Jobs, and CronJobs. Create workloads from a client, in the console, or by using a template. Configure environment variables, health checks, data disks, and logging as needed.

Application scaling

ACS supports manual scaling, Horizontal Pod Autoscaler (HPA), CronHPA, and AHPA.

Storage management

Volumes in ACS clusters are managed by the CSI plug-in and support Elastic Block Storage (EBS) and Apsara File Storage NAS (NAS).

Network management

ACS provides stable, high-performance container networks by integrating the Kubernetes network model, Virtual Private Cloud (VPC), and Server Load Balancer (SLB).

Security and O&M

Category

Feature

Description

Observability

Monitoring

Managed Service for Prometheus is integrated with ACS and enabled by default. Built-in dashboards and performance metrics let you monitor Kubernetes clusters, pods, and applications across multiple dimensions.

Logs

ACS is integrated with Simple Log Service for collecting and viewing application logs, pod logs, and cluster logs.

Alerting

Configure alerts to manage cluster exceptions based on various metrics and scenarios.

Cluster inspection

Cluster inspection

ACS provides the cluster inspection feature to automatically scan for potential risks in clusters.

Diagnostics

ACS provides the diagnostics feature to diagnose pods, Services, and Ingresses.

Security center

Audit

ACS generates API server audit logs to record and trace daily user operations.

Limits

Take note of the following limits before using Container Compute Service (ACS).

  • ACS clusters do not support DaemonSets. Use sidecar containers instead.

  • You cannot specify HostPath or HostNetwork in pod manifests.

  • ACS clusters do not support privileged containers. Use a security context to add capabilities to a pod.

  • ACS clusters do not support NodePort Services or session affinity.

  • ACS clusters do not support the China South Finance, or Alibaba Gov Cloud regions.

References

  • For more information about how to activate and use ACS, see Get started with ACS.

  • To view the release notes of ACS, see Release notes.

  • For more information about solutions used in different scenarios based on ACS, see Best practices.