Collected Data, now sizing VMs & Host.

[Greenwich789]

I'd like to check to see if the way I've started to collate data on our VMs is valid and therefore worth perusing.

Using powercli I'm pulling back 5 min data of all VMs in our environment, The script runs at the end of the day and pulls back various metrics (cpu.usage.average, cpu.ready, mem.active.average..etc). They are then processed and I produce a day summary of each VM which I can then aggregate into weekly, monthly etc. As I have 5 minute data I'm able to see the finer peaks and troughs that might be missed if the longer periods were used.

An example of a days data is attached

B - Num CPU assigned

C - Average CPU usage for day,

D - Num of CPU user can be resized to (Cpu * Avg)/100,

E - New potential CPU size of VM (column D rounded up)

F - Max CPU used

G - Num of CPU VM can be resized to based on days Max

H - New Potential CPU size of VM based on days Max (column G rounded up)

Second table row shows the SUM of the column.

Question1

Now, is it valid for me to take the sum from column H (or perhaps column E) and use that to size a host. So, based on 47, I would need a 48 (rounded up) processor (inc hyperthreading) host if I wanted to move these VMs onto another host. (leaving aside overhead).

Question2

The VMs are oversized in the memory department too. If they were reduced, would there be any effect at all on CPU if Ram were reduced?

Question3

Anything obvious I'm missing or should be including?

Appreciate feedback. Apologies if i've not been clear, feel free to ask for clarification.

[weinstein5]

Welcome to the Community -

Q1) Sizing your environment so that the number of physical cores to the the number of virtual CPUs will deifinitely be overkill - the current ratio of vCPUs to physical CPUs is 8-10 vCPUs to a physical core - so an ESXi hosts with 16 cores should easily handle 100-160 virrual cpus assuming of course there are no special workloads with extreme resource requirements

Q2) Reducing should impact CPU requirements

Q3) you are taking a good approach -

[Greenwich789]

Hi - thanks for your comments.

Now here is my confusion.

I'm aware that generally I would be hoping to get a consolidation ratio that was more in the region of 1:4 - 1:6, 1:1 is hardly an achievement!  The numbers that I present in the original post are taken from a host that is almost exclusively running only those hosts (there are a few VMs also running that are not particularly heavy). When I look at the days utilization via the vSphere Performance tab, I'm seeing values of around 80% average usage of the host and just a few percentage points higher for the max.

The reason for posting was that I wasn't expecting the host to be nearly maxed out. (The host is a Xeon X5675 (12 CPU x 3.07Ghz) Hyperthreading is enabled. Its a host running linked clone Win7 64bit View desktops)

I understand that each environment is different and each one will achieve differing levels of consolidation. As I'm seeing such low numbers, I was concerned that I had missed something crucial. These calculations will be used to move forward to a new environment so its important that all bases are covered.  If I haven't missed anything, it may just be that these users (these are thought of not high usage within our financial company) are heavier than expected.

Doesn't add up though, does it?