Hello,
We currently have several vsphere hosts running ESXi 4.1 Update 1 with 128 GB of RAM each.
One of the hosts shows 93% utilization in vCenter, but when viewing the performance chart shows barely over 10% utlization. This is confirrmed in the vCenter Operations view showing 14% utilization.
This led me to this article:
http://www.yellow-bricks.com/2011/04/29/which-metric-to-use-for-monitoring-memory/
There is basically no swapping going on. I also ran a chart in vCenter Operations showing 1% memory contention at the most.
So the question is is there much of a performance hit if the host is totally consumed while only a small percentage of memory is being used?
Would there be much of a benefit of doubling the RAM in the host. (It is a DL580 G7 and is upgradeable to 256 GB ) As of now I dont see a benefit if I doubled the RAM when such a small percentage of memory is active, but I could be wrong.
Do you guys have a general rule of thumb for a ratio of allocated ram to physical RAM? Of course this depends on how much is actually active.
We have a lot of SQL and Exchange VMs which only use a fraction of memory but the argument is if we dont give them X amount of memory Microsoft won't support them. These resources add up big time when you have a bunch of SQL clusters (active/passive).
I am like a lot of other vmware admins where getting people to understand they dont need X amount of resources and only Y is being used is like pulling teeth.
What memory metric do you monitor the most and howmuch memory overallocation will you do for an ESX host?
Thanks for any input on this.