Container Service for Kubernetes:Recommended ECS instance specifications

Cluster specification planning

Using many small ECS instances to create an ACK cluster can cause the following issues:

• Network limitations: Small worker nodes have limited network resources.

• Resource capacity: To ensure cluster stability and reliability, the system reserves some node resources, such as CPU, memory, and disk, for cluster management and infrastructure components. This reservation can significantly reduce the available resources on small ECS instances, which affects cluster performance and availability. For more information about the node resource reservation policy of ACK, see Node resource reservation policy.

• Resource fragmentation: When resources on a small ECS instance are allocated to a container, the remaining resources can become fragmented and unusable. These leftover resources cannot be used to create new containers or recover failed ones, which leads to resource waste. For example, if a node can allocate CPU only in whole units and an application requires only a small amount of CPU, the remaining CPU resources on that unit may be wasted.

Using large ECS instances provides the following benefits:

• Improved network performance: Large instances have high network bandwidth, which is ideal for high-bandwidth applications. In addition, more containers can communicate within a single ECS instance, which reduces cross-node network traffic.

• Efficient image pulling: On a large instance, an image is pulled only once and can then be used by multiple containers. In contrast, a cluster with many small ECS instances requires the same image to be pulled multiple times. This process takes longer when you scale the cluster by adding new instances and delays the response.

For more information about how to select ECS instance types, see the following sections.

Select worker node specifications

• Use node specifications with a minimum of 4 CPU cores and 8 GB of memory.

• Calculate the total number of CPU cores required for daily use and determine the availability requirements for the cluster.

  For example, assume a cluster requires a total of 160 CPU cores and must tolerate a 10% fault rate. In this case, you can select at least 10 ECS instances with 16 CPU cores each. The peak operating load must not exceed 144 CPU cores (160 × 90%). If the required fault tolerance is 20%, you can select at least 5 ECS instances with 32 CPU cores each. The peak operating load must not exceed 128 CPU cores (160 × 80%). This configuration ensures that if one ECS instance fails, the remaining instances can still support your services.

  If your cluster's daily scale reaches about 1,000 CPU cores, you can use ECS Bare Metal Instances. For more information, see Scenarios and benefits of ECS Bare Metal Instances.

• Determine the CPU-to-memory ratio, such as 1:2 or 1:4, based on the resource requirements of your pods. For memory-intensive applications, such as Java applications, consider using instance types with a 1:8 ratio.

• To maintain service stability and ensure accurate resource scheduling, do not mix GPU-accelerated and non-GPU instance types in the same node pool.

• Using persistent memory-optimized instances

  Worker nodes that are persistent memory-optimized instances, such as re6p, use a hybrid memory architecture that includes both regular memory and persistent memory. To implement persistent storage, see Non-volatile memory volumes. For more information about persistent memory-optimized instances, see Instance families.

Select master node specifications

When you create an ACK cluster, core components such as etcd, kube-apiserver, and kube-controller run on the master nodes. For production ACK dedicated clusters, you must select appropriate master node specifications to ensure cluster stability. The required specifications depend on the cluster size, and larger clusters require higher specifications.

You can use small ECS instances for personal testing and learning. For production clusters, select master node specifications from the following table to maintain the master node load at a safe level.

Scenarios and benefits of ECS Bare Metal Instances

ECS Bare Metal Instance is an innovative computing service developed by Alibaba Cloud based on state-of-the-art virtualization 2.0 technology. Virtualization 2.0 endows ECS bare metal instances with the elasticity of virtual machines (ECS instances), the performance and features of physical machines, and full support for nested virtualization.

Unsupported ECS instance types