Recommendation for percentage RAM free

[philRAL]

We currently have a 5 node proxmox 8 cluster using Dell R650 servers, each with 768GB RAM and dual socket 28 core CPUs.
We are in the process of migrating VMs onto the cluster from VMware (about 150 VMs so far) and are currently at around 30% memory usage on each hypervisor. CPU usage is low (less than 5%).
At what point does it make sense to add more hypervisors? We usually keep one hypervisor free of VMs for maintenance purposes. The workload profile is generally predictable and stable, no sudden bursting etc.
Where would folks recommend we set as a memory usage limit before expanding the cluster? Would 75% sound about right or higher?

 

[Eduardo Taboada]

IMO, there are a lot of factors dealing with RAM, because in certain conditions you need to raise the memory of some virtual machines, if you are running CEPH, you can use the empty hypervisor to provision VM (leaving room to move VM's out of this hypervisor for maintenance and updates) so you can have less memory used in each node.

 

[SteveITS]

Obviously having one node offline is common (e.g. updates/reboot). Do you need to allow for a second outage/failure during maintenance? Divide total usage by 3 instead of 5.

 

[philRAL]

Obviously having one node offline is common (e.g. updates/reboot). Do you need to allow for a second outage/failure during maintenance? Divide total usage by 3 instead of 5.

yes that is a good point.