RawDeployment and Serverless comparison page #489
Description
Describe the change you'd like to see
Docs are missing a page where all the differences between KServe's RawDeployment and Serverless mode for ISVCs. This could include features or limitations of each mode, plus some guidance on what to pick based on use cases. There is some information in the RawDeployment installation guide but I think this is far from complete.
Additional context
Apart from the information in the aforementioned doc page, I managed to gather also the following information from other doc pages:
Autocaling happens with Kubernetes Autoscaler(HPA), in contrary to Knative serving, it
- Does not support scale to zero.
- Supports CPU-based autoscaling.
RawDeployment with KEDA can be used to scale using:
- CPU Utilization
- Using Metrics via Prometheus
- Using Metrics via OpenTelemetry
RawDeployment supports Multi-node functionality, where auto-scaling is not available.