You're correct that it depends on various factors. But lets look at the rough estimations:

For an OpenShift Cluster with high availability across 3 locations will require 3 Master node, and Each Master node requires at least 4 vCPUs(as per OpenShift documentation). So the Calculations for vCPUs for Master nodes:

3 Master Nodes * 4 vCPUs per node = 12 vCPUs

• Note: This is just an estimation, and the actual vCPUs could vary based on other factors like load balancer, database queries, etc.

For Red Hat 3Scale, there are no guidelines on vCPUs requirements per RPS, but here's an estimate for 70 - 100 RPS:

• For lightweight API request (no heavy processing) requires at least 0.5 - 2 vCPUs.
• For moderate API request (database queries, basic calculations, validating data) requires at least 2 - 4 vCPUs.
• For heavy API request (large file upload, complex calculations, quering large database) requires at least 4 - 6+ vCPUs.

Since 3scale's API Gateway handles traffic, routes requests to appropriate backends, applies policies (e.g., authentication, rate limiting), and returns responses, for 70-100 RPS, we can assume you'll need about 1-2 vCPUs for simple usage.

So, if you’re working with lightweight API requests (70 RPS), and you're using 3scale with basic policies or authentication, you’ll likely need at least 2 vCPUs. If you’re applying more complex policies, rate limiting, or integrations with backend systems (like databases), you may need 2-4 vCPUs.

Note:

• 1 Node = 1 vCPUs
• While scaling application we need to increase vCPUs.
• If we use any monitoring, logging, service mesh, Kafka or redis services, we may need to increase vCPUs.
• A single database instance may need 2 - 8 vCPUs, still it depends on the scale of the workload.
• OpenShift Cluster across 3 regions with monitoring, logging, CI/CD, 3Scale and Database, you could easily end up needing 50 - 100 vCPUs depending on the specific workload.

Start with 1 - 2 vCPUs for simple usage. Scale to 2 - 4 vCPUs if you have more complex API policies, backend processing. This is just and approximation, and you may need to monitor performance in real world environment, so adjust as needed based on actual load and service requirement.