I’ve read a fair few articles on this (and the changes post 6.5), and I’m either enlightened or baffled. Not sure which yet
Our hosts have 2 sockets, each with 16 physical cores, 128GB memory (4x32GB) in total for the host
We have a VM (SQLServer) which is using 10 vcpu – so that fits in a single pNUMA node, however its using 80GB memory. I think that if it was using only 64GB then it would fit with a single pNUMA, the fact that it is using 80 means that it is needing to use remote memory. which isn’t ideal
I was actually think of adding another 64GB (2x32GB) to the host giving 196GB, so 98GB per pNUMA node.
still trying to decide whether (we have Gen 10 DL380 with Intel 6130 Chips) whether that should be 3x32GB sticks per chip, or 4x32GB against one socket and 2x32GB sticks on the other socket
Does any of that make sense or have I lost whatever grasp I had on it at any point today!
To see the full content, share this page by clicking one of the buttons below