I have bought the e-book Cluster Deepdive, and it is very technical.
I may have missed this, or misunderstand it in the book, or it went over my head. But if I am using a Percentage Based HA admission policy, and my hosts are dis-similar. How would calculation be?
An example, though not mine, would be say you have cluster of 4 hosts. 1 Host is 128GB, 1 Host is 256GB, and 2 Host are 512GB. How would you calculate a percentage to reserve? I know if the hosts are all the same it is an N-1 formula. I must have missed this in book.
As I wrote this I think I understanding better. I need to stop thinking about protecting a single host, like one of the admission polices. But more about protecting a certain amount of memory that can be spread around any of the hosts?
With dissimilar sized hosts you want to think of it in terms of your largest host and determine the % from there. So taking your example of 4 hosts and calculating for memory:
Host 1 128GB
Host 2 256GB
Host 3 512GB
Host 4 512GB
Total memory: 1408GB
Taking one of the largest hosts (512GB) and determining what % that host represents of our total resources (to calculate for n+1) we get 512/1408 = .364 or 37%. Setting your percentage based admission control policy setting for memory to this will ensure you have enough resources to restart all VMs if you lose any of your hosts.
Does this answer your question?
This will work only if all of your VMs have CPU and memory reservations configured, which I doubt they do. As percentage based admission control ensures resources are available only for VMs with reservations, other VMs are not considered in calculations.