Kubernetes nodes are essentially the core building blocks of K8s clusters. How do you determine the right number of nodes for clusters? Read on.
Kubernetes nodes are essentially the core building blocks of Kubernetes clusters. As a thumb rule, people say that more nodes in a cluster equal higher availability and performance. But, that doesn't mean that enterprises should just add an excessive number of nodes on their K8s clusters. Such addition could lead to resource and cost wastage.
The easiest way to determine the number of nodes is to assess performance and availability requirements in advance and deploy nodes accordingly. Here is a deeper look at the various parameters to consider in deciding the number of Kubernetes nodes to have in a cluster.
Kubernetes Nodes and High Availability
The number of Kubernetes nodes in clusters has a direct relationship with the workload availability of the environment. For example, when there are only a few nodes to manage the workload, failure of a couple of them can lead to insufficient nodes to store the Kubernetes pods. Another example is when a master node (which manages all other 'worker nodes') fails, the entire cluster could collapse.
How do you solve these issues? Although the number of nodes can vary, it's safe to say that there should be at least two master nodes and a dozen or so worker nodes to ensure good availability and redundancy. The ideal number should be determined according to organizational needs.
Nodes and Performance
On any given node, parameters such as compute, memory, etc., can vary according to the creating server's hardware profile and specifications. Teams need to determine how much resources each node delivers to the respective clusters. Therefore, in performance, the amount of resources plays a more significant role than the number of nodes. A few high-powered nodes can deliver better performance than a large number of low-powered nodes.
It is always safe to keep a buffer above the performance you determine - say, about 20% - if there is a need to balance the cluster performance against sudden failures or peaks.
Physical or Virtual Machines
Organizations must decide whether the Kubernetes nodes are based on physical, dedicated servers, virtual machines (VMs), or a combination of both. Compared to physical servers, VMs pose a greater risk to the nodes and clusters. Having a dedicated physical server for each node lessens the chance of multiple nodes failing together. On the other hand, it is slightly more expensive than VMs.Therefore, the ideal approach is using a mix physical of servers and VMs for Kubernetes nodes.
