We're not that advanced here yet.
All the design and approval was done prior to me joining the company, so we have 2 hosts, one at each site running their own set of VM's on local HDDs. This is far from an ideal VMware setup, I know. I'm trying to push the company to invest in a SAN and to bring the 2 ESXi hosts to the same site and run in a HA cluster for resilience.
20k VMs? Hmm. I think it's definitely time to look at Orchestrator to automate that verification.
The KB you refer to will apply to VMs that have a limit set which is rare. Mostly explicitly configured by the administrator. But a good point to verify. In fact it was a case used to demonstrate the use of vC Ops to troubleshoot a performance issue in a VM.
I want to reiterate that vCenter reports on active memory, which is the memory that has been recently used. The guest OS reports used memory which is the memory allocated to all processes whether it is being accessed or not.
What this value means is that if the host were to swap out memory pages belonging to that VM it would first select pages that are not active. If this were to happen the impact on performance would be minimized (but not eliminated) because those pages are not actively used.
In other words it is a good indicator for right sizing your VMs. Mind you, you still have to take into account the minimum requirements for the OS and application, but the latter is often exaggerated by the application provider.
The article actually indicates that it may be a carry-over from previous versions and the VMs I saw the issues with were originally on ESX 3.5 (and looking at the Tools version it is still on 3.5, so there’s another update to be done soon ☺). It was also the case that I wasn’t seeing any persistent balloon on the VMs that would correlate to what was being indicated as being used in Windows either.
Not having other hosts to vmotion to isn’t the end of the world – we have many outlying sites with single hosts and local storage where it was just a business decision to take the hit if maintenance needed to be done on hosts, so VMs had to be taken down.
To go back to the original issue though – what does Task Manager show in respect of memory used by processes, compared to what is reported by the Performance tab? That was another thing I noticed in that the Processes were using (at a rough calculation) around 8GB (which correlated to what vCenter said was being consumed), while Performance tab reported using 14GB. That could again point to a potential issue with balloon driver.
Looking at one of my mahcines right now I see the following:
OS - Processes using approx 650MB
OS - Performance, Physical memory usage, 755MB
vSphere reports 1.5GB active
I am seeing no ballon driver activity on the VM in the vSphere console either.
I have similar kind of issue.
Task Manager of Windows Server shows 15GB RAM utilizatio out of 16GB.
Here i Checked the process TAB where sql service is using around 15GB.RAM out of 16GB.
But when checking the Vcenter Server
VM active memory shows 4GB used.
Consumed memory shows utilization around 15GB in Vcenter Server Console.
VM is running slow
Can any body guide how i can troubleshoot the issue?
We have same issue.
In our SQL have set up minimum memory usage to 60GB. Granted memory is 64GB.
In vCenter Performance monitor's Memory Active counter's value is only 15GB, but the OP system show 62GB usage.
Anybody have idea, why??