Container Service for Kubernetes:Overview of ACK One

Cluster types

ACK One organizes its capabilities around three cluster types:

Capabilities

Manage clusters from one place

• Connect clusters from any provider or location to a single console and API surface.

• Centrally enforce security policies, access controls, and configuration inspections across all clusters.

• View health and cost metrics for all clusters in one global monitoring dashboard.

Scale resources on demand

• Burst workloads from on-premises clusters to Alibaba Cloud by adding Elastic Compute Service (ECS) instances or ECI to external Kubernetes clusters.

• Use the ACK scheduler for advanced scheduling: gang scheduling, topology-aware CPU scheduling, and ECI-based scheduling.

• Accelerate data access and reduce bandwidth usage with ACK Fluid distributed cache in compute-storage decoupled environments.

• Scale cloud resources automatically to handle traffic fluctuations, or set scheduled scaling to improve cost-effectiveness.

Protect and recover applications

• Back up and restore applications and data across regions or from data centers to the cloud using the Backup center—available out of the box with no additional setup.

• Set automated backup policies and restoration policies to keep applications continuously protected.

• Build active geo-redundancy across three data centers in two zones for business continuity.

Distribute applications across multiple clusters

• Host open source ArgoCD in ACK One and distribute multi-cluster applications through GitOps.

• Apply different configurations per cluster while deploying from the same Git repository.

• Run jobs across multiple clusters on a schedule.

Manage traffic at the fleet level

• Route north-south traffic across clusters using MSE cloud-native gateways.

• Create multi-cluster Services to manage east-west traffic.

• Configure Global Ingresses with Layer 7 routing rules based on weights and pod replica counts, with automatic fallback.

Run AI and big data workloads

• Quickly deploy large numbers of enterprise-class products or components verified by Alibaba Cloud in Kubernetes clusters to enhance security, improve scheduling efficiency, and accelerate AI and big data computing.

• Manage AI training jobs, resource quotas, and observability from a unified interface.

• Improve GPU utilization by approximately 300% with GPU sharing.

• Accelerate distributed training with compute-storage decoupling and cross-cluster scheduling for Spark, Kubernetes, and TensorFlow jobs.

• Enable intelligent CPU scheduling and non-uniform memory access (NUMA) awareness on ECS Bare Metal instances.

Run large-scale workflows cost-effectively

• Pay only for data plane usage—Argo workflows control planes are free of charge.

• Use preemptible instances to reduce compute costs further.

• Handle thousands of concurrent workflows and tens of thousands of concurrent computing tasks.

• Trigger workflows automatically from Git, Message Service (MNS), or Object Storage Service (OSS) events.

• Achieve more than 20 Gbit/s aggregated read bandwidth with distributed cache across regions.

Use cases

Connect on-premises clusters and scale to the cloud

Register an on-premises cluster to connect your data center to Alibaba Cloud. During traffic peaks, ACK One scales resources and applications to the cloud to balance load, so your on-premises infrastructure handles only baseline demand.

Extend on-premises clusters with Alibaba Cloud services

Bring Alibaba Cloud observability, security, and microservice governance capabilities to clusters deployed in data centers or third-party clouds:

• Observability: Collect logs, monitoring data, and events with consistent O&M experience across environments.

• Security: Enable auditing, security inspection, node risk detection, and policy governance.

• Microservice governance: Use Service Mesh (ASM) and Microservices Engine (MSE) for traffic control and service governance.

Implement disaster recovery across hybrid cloud, regions, or zones

• Back up stateful applications and data across regions or from on-premises to the cloud.

• Schedule automated backups and define restoration policies to meet recovery objectives.

• Build an active geo-redundancy architecture with three data centers across two zones for Kubernetes-native business continuity.

Accelerate AI and big data workloads

• AI algorithm development: Manage AI jobs, quotas, and observability from one console.

• AI training: Use topology-aware scheduling, compute-storage decoupling, and cross-cluster scheduling for Spark, Kubernetes, and TensorFlow jobs.

• AI inference: Improve GPU utilization by approximately 300% with GPU sharing, with autoscaling across cloud and on-premises resources.

• Intelligent CPU scheduling: Run NUMA-aware workloads on ECS Bare Metal instances for latency-sensitive jobs.

Distribute applications to multiple clusters through GitOps

Use a multi-cluster Fleet instance with hosted ArgoCD to deploy applications from Git repositories to multiple clusters simultaneously:

• Developers need only Git repository permissions—no direct Kubernetes cluster access required.

• Apply version control, change approval, code rollback, and audit logs to every deployment.

• Keep applications in clusters continuously synchronized with the state declared in Git.

• Deploy the same application with different configurations to different clusters.

Implement zone-disaster recovery with multi-cluster gateways

Route traffic intelligently across clusters to reduce costs and improve resilience:

• Use multi-cluster gateways to schedule north-south traffic based on availability and cost.

• Create Global Ingresses with Layer 7 routing rules controlled by weight and pod replica count, with automatic fallback when a cluster becomes unavailable.

Orchestrate large-scale jobs and complex workflows with Argo workflows

Run simulation, scientific computing, data processing, and continuous integration workloads on a managed serverless Argo workflows control plane:

• Use resources across multiple regions and zones.

• Reduce costs with preemptible instances and pay-per-use data plane billing.

• Decouple computing and storage with distributed cache to accelerate job execution.

Next steps

• Overview of registered clusters

• Fleet management overview

• Overview of Kubernetes clusters for distributed Argo workflows

• Backup center

Contact us

If you have questions about ACK One, join the DingTalk group 35688562.