The resources calculator should be created from empirical observations, and based on actual/current cluster metrics, provide estimated ranges of values (CPU/memory/storage cost at least). It must include some variations, e.g. with sampling 1 / 50 / 250, with / without Loki. Using Kafka or not should follow our recommendations.

It should also provide an estimation of the latency overhead on traffic.

Metrics to consider for calculating the estimation are: (can be refined)

• current cluster bandwidth
• number of pods
• number of nodes
• ...