I'm looking to see the total amount of memory used by all virtual machines and by the ESXi hosts itself added together on the performance charts. Do I want consumed, active or granted as my metric?
AFAIK if you are trying to understand the VM memory usage of the VMs in the Infra one should consider Active Memory usage because its what the VMkernel believes is currently being actively used by the VM. I have used this metric in the past while doing CP and it gives more realistic data. In short Consumed Memory = Granted – TPS – Memory Overhead (TPS & Memory Overhead whereas Granted Memory = all physically mapped pages. For more details please review the following articles (& comments) by Duncan Epping
About VMware Virtualization on NetApp
Thanks for the input. Right now I'm attempting to measure the memory use on the entire ESXi host by all VMs, rather than by a single VM. Because I am using statistics level 1, active memory is not viewable for anything but real-time at the ESXi host level. In light of that, which metric is best for viewing the sum of all memory used by all VMs by the ESXi host over the last week?
Most of the counters are not available beyond 1 hour. vSphere is not a performance monitoring tool, it's more like a platform administration tool. You should be using the bundled vCenter Operations (the edition bundled with vSphere is free. Certainly, it consumes resources too) for performance management and troubleshooting.
I'd watch for Contention first, then Utilization. This is because high contention is bad, regardless whether you have high or low utilization. On the other hand, if you have high utilization, but low contention, that means your investment on the infra is being used well 🙂 You may need to buy more infra, but that's a Capacity Management discussion, not Performance Management (I agree they are the same topic on 2 different time scale).
For utilization, I'd watch for the following:
- for VM: Usage (which is Active)
- for ESXi: Usage (which is Consumed) and Active
I'm tracking Consumed at ESXi level as it shows what was active before.
I won't worry much about Granted once I track Consumed, as Consumed = Granted - Shared.
Hope that helps