I'm building a kubernetes based developer environment where each user is able to launch a set of services (deployments, sts ..) into a namespace in a cluster. Each namespace is an environment and we have a single cluster for multiple users.
Based on the current usage trends of developers, I'm trying to project how much the cost would be for 1000 developers using this environment. My idea is this,
- Take the # of users per day who are actively using the cluster
- Get the cost of the cluster per day
- Build a linear regression model between active users and cost
Actualy I'm already able to do this but the R^2 which is usually used to determine the quality of the model is 0.64. And this is because the usage pattern for the same # of users across different days might vary.
So I'm considering going more granular to per hour (for which I have the data) which would be extremely accurate but I'd lose the ability to estimate per day/month. So I'm seeking feedback on the current model or suggestions on a better approach.
