Many of my hosts often report memory congestion, and I am trying to find out what vmturbo considers "congestion." In one cluster my hosts average 75% memory usage and there is no congestion reported, while another averages 35% and does report it.
So I tried an experiment. I have a relatively new ESXi cluster that I am still migrating VMs to, and utilization is at about 20% (memory). After applying patches, half of the hosts currently have no VMs on them. The other hosts are at 40-52% memory and are all reporting memory congestion in vmturbo. I put the empty hosts into maintenance mode, waited a little while, and the remaining hosts are no longer reporting memory congestion.
So, is congestion a metric, or just a flag indicating there could be better VM balancing across the cluster? How do others use this information and what do you do about it?