CPU and RAM Overcommit ratio

[vharry]

Hi All, 1. Is there an out of the box metric in vRops 6.2 for the RAM Overcommit ratio for a cluster for both powered on and powered off vm's? 2. Is the CPU|Demand|Consumption ratio based on the powered on vm's or all vm's or the equivalent of vCPUs for the amount of CPU Demand in MHz? thanks in advance...

[sxnxr]

You are talking about two different capacity models, Allocation and demand.

The ram over-commit on a cluster you set in the policy. It will take into consideration powered on and off as this is the allocation model (you cant use over commit ratios in demand model as far as i know). You can see if you have over-allocated the over-commit in environment --> Select object --> analysis --> Capacity remaining. You will need to work it out between the Peak value and usable capacity. Here you can also see what your over commit ratio is. This is also where you can take advantage of Custom profiles to show you how many of your 2vcpu x 16gb ram will fit into your cluster.

the CPU|Demand|Consumption ratio is basically number_running_vcpus divided by corecount_provisioned so it only counts powered on VMs as it is part of a demand based model (Powered off VM dont demand anything apart from storage). It is based on vcpu counts and physical core counts and not MHz

[vharry]

"You will need to work it out between the Peak value and usable capacity* .... I think this is part of my problem in that I can't tell how many GB of RAM has been allocated (configured from vm properties) to all virtual machines (powered on and off), I need this value to compare against the total memory to work out the current ratio I am running at. For example one of my clusters have vSphere replicated vm's that I need to take into account as they may start up at anytime and I would consider them as part of used capacity. I thought it would make sense to have the following metrics; CPU|Allocation|consumption  - Powered On and Off Overcommit ratio RAM|Allocation|Consumption  - Powered On and Off Overcommit ratio RAM|Demand|Consumption    - Powered On Overcommit ratio

[sxnxr]

If you look under environment --> (select object) --> analysis --> Capacity remaining you will get

This will show you the over commit, the peak and the remaining. In the above example i am not over committed. I have a total capacity of 9.47 TB (simply adding up all the memory in all the hosts) I have a 10 % reserved for a buffer so i have a usable of 8.52tb (take 10 % off the Total gives me my usable) My peak value is 7.35TB (This is simply the adding up of all the memory allocated to all VMs (This i think counts powered off vms as well because they are still allocated memory). So you take the peak value away from the usable and you get the remaining.

The reason i say you will need to work it out is because when your usable drops to say -1tb (over allocated memory to the vms by 1TB passed the usable) this will only display 0 and not a negative number. As you can see from the below my peak is higher than my usable by approx 780GB so this is the amount over allocated.

It should be easy to see if it takes into consideration powered off vms by creating a view listing all the VMs and the configured ram and adding it up then compare to the peak value. It may not be exact as the above and below are based on peak not latest.

If all works you could use the memory capacity remaining metric to show if you are over allocated as you can add that metric to a view and create dashboards and reports. Might also be good to include the memory peak as well co you can see of the remaining is 0 how much over you are.

Hope this is what you are after

[vharry]

I understand that the "configured" overcommit ratio is shown under the capacity remaining, what I am looking for is what ratio it is currently running at. for example your first screenshot shows 5:1 cpu ratio and considering you have about 57% remaining, the current ratio is roughly 2.5:1. You have shown what the value is of the resource that is remaining but I want to see the relevance of the current ratio to the "configured" ratio, then perhaps look at winding it down or up to gain efficiency. And more so for vRAM.

One thing you have pointed out is the peak value which I believe maps to the "(RAM or CPU)|Allocation|Computed Demand" metric for either CPU or RAM. I can then create a supermetric to calculate "computed demand"/"total capacity" which should give me the current ratio. However it's still not clear whether computed demand is including powered on or off as I am getting mixed results. I tend to need both as some clusters need to take powered off into account whereas others don't, based on business rules.

[sxnxr · Accepted solution]

If you look under density it will show you the current ratio (analysis tab for the object)

Also you can graph it with the below

As for the powered off and taking them into consideration depending on cluster you might be able to use a super metric but i have not done to much with them. I know in 6.3 they introduced a lot of new stuff like if statements so that might be able to add up the allocated VMs if they are powered on.

[vharry]

thanks sxnxr, I guess the two metrics I am looking for are;

cpu|demand|consumption ratio

memory|demand|consumption ratio

what has really confused me is that these two metrics calculate the Allocated resources to Powered On VM's so I think they should be under Allocation and not Demand. I had the understanding that these were calculating the ratio based on the MHz demand in number of VCPU's. Perhaps that is that the optimal version of this metric is.

the other variations of this metrics would be nice such as those taking powered off vm's into account and also taking redundancy/buffer into account (such as total capacity - buffer%).