How to measure memory overcommittment threshold?

[tmilner]

In short I'm trying to determine what the total memory requirement is of all running VMs. In other words, how much memory must my esx server have before it becomes overcomitted and must starting freeing up memory.

My current algorithm is to make a performance query and sum the counter "Memory Consumed (Maximum)" for all VMs. But I'm confused about the host counter "Memory Overhead (Average)". Should I retrieve that and add it in? Or is it the average overhead per VM?

And mostly, is this overall algorithm correct or am I missing something? TIA.

Tom

[tmilner]

Amazing!

All the clatter about how VMware's memory overcomittment saves money vs. Hyper-V... how putting esx server in an overcommitted state causes unwanted overhead (balloon memory, sharing of pages, swapping, etc.)... and yet no one knows how to calculate the threshold?

I guess this "overcomittment" thing is like a beer stein with a big hole in the bottom... Interesting to discuss, but not very handy to work with.

Tom