How to select computing resources for scale-out cluster nodes?

Question

Scalability

With horizontal scaling, we add more nodes to the cluster to handle more load. But as before, we can choose how much power (CPU, RAM) each cluster node will have.
When the cluster is managed by Kubernetes, it is very easy to set a limit on CPU and memory for Pods. How to choose the optimal CPU and memory size for cluster nodes (or Pods in Kubernetes)?
For example, in a cluster there are 3 nodes with 1 vCPU and 1GB RAM each. To process more requests there are 2 options:

• Add 4th node with 1 vCPU and 1GB RAM
  
• Add capacity to each of the 3 cluster nodes (for example, 2 vCPUs and 2GB RAM)
  

The obvious solution seems to be to compare both options in terms of throughput and cost and choose the best one. Are there more advanced approaches to calculating the optimal size of computing resources for the nodes of a scale-out cluster?

Answer 1

The answer has to do with performance metrics like latency and throughput .
Latency - the time interval between sending a request and receiving a response.
Throughput - request processing speed (requests per second).
Latency affects throughput. More latency - less throughput.
If a business transaction consists of consecutive calls to several services that cannot be parallelized, then the computing resources for the cluster nodes should be chosen in such a way as to achieve the required latency, since adding additional service instances (horizontal scaling) will not have an impact on latency.
Adding additional service instances increases throughput (provided there are no bottlenecks).
That is, choose the CPU and RAM in such a way that the response time of the service is optimal, and to process more requests at the same time, add new service instances.