We are monitoring kubernetes pods through the Application only monitoring mode and I was curious if the host units were calculated using memory requests or limits of the pod? What exact metric of the pod's memory is used in determining the cost?
It's the maximum memory available for the pod - the memory limit.