Simon has done work in the past: https://docs.google.com/document/d/1T2fgc10XW7MwD3tQhbLl7DqI8BoY0Qk39FKIwzF6D4g/edit#

This investigation has however compared the default CMO deployment (15d retention) with agent mode. Hypershift if potentially interested in running prometheus in server mode with a small retention period. it would be good if we had a clearer picture of the potential resource usage saving between a smaller retention and agent mode.