I am new to VRops, trying to figure out the best way and metrics to monitor the guest memory please
some recommend this is best to be done from the guest itself using perfmon or resource monitor.
I am aware VRops can use End point agent on guests to get more accurate readings.
What metrics would you recommend?
I am trying to understand of using metrics for optimals VM RAM sizing.
Many thanks all
It all depends on the version of vrops, ESXi and VMware tools you are running.
If you are running
vROps 6.3 or later
VC 6.0 U1 or later
ESX/ESXi 6.0 U1 or later
VMware Tools from VMware ESX/ESXi 6.0 U1 or later installed in the VM
then you should get the in guest stats with out the need for the eops agent
then look for the guest|mem.needed_lates metric. this is what perf mon would report. One thing to remember is vrops collects every 5 mins and perf mon is real time so there may be a slight difference if comparing the two
If you are not running one of the above min requirements then you will need the eops agent but that introduces more complexities
Guest memory monitoring is and has always been a very tricky business. While, yes, there are metrics in vROps like those mentioned, they are VM-level metrics, and no matter how fine-grained they are they only attempt to "guess" at what is going on within the guest. The only true way to understand how applications are consuming and using memory is by actually looking at application-level metrics and pairing them up with in-guest metrics (via an agent like EpOps or something else). When you look at guest memory from outside of the VM, it's kind of like figuring out how many people are inside a van with tinted windows. You can pick it up and shake it around to listen for sounds, but until you peek inside you really are just taking an educated guess.
Many thanks all
I found the metrics on guest Perfmon , VM CPU and VM RAM, i believe these are using vmtools and are best way to monitor guest resources.
What metrics do you use for ESXI host ram monitoring? Consumed RAM from Vcenter is monitoring every 20 seconds, what are the better ones from VRops? I guess ESXTOP mighnt be better option as real time?
I am trying to find a best way to monitor and size the VMs, I understand the NUMA concept and ideally VM should stay withing same NUMA domain rather then crossing in between.
If the Blade has 2 sockets x 12 cores, i understand the single NUMA domain would be 12 cores or 12 VCPUs? so VM should not be allocated with more then 12 CPUs in order to stay in the same NUMA domain?
in this example it would have 24 cores in total or 48 Logical CPUs if Hypertreading on, does this mean that max VCpus on single ESXI host should ideally be 48 vCPUs (same as Logical CPUs)? that would be ratio 2:1?
Many thanks all for your responses