ESXi 6.5 Not Providing Sufficient Processing Power to Single VM

[KMarn]

Hello, I am having an issue setting up a system in ESXi and would really appreciate some tips on how to get over the hump. I think what I am trying to do is a little bit unconventional, and I've had a hard time finding specific guidance through Google on it. Please note that this is not my field of expertise, I've just been kind of thrown into getting this working. Please forgive me if I use confusing or incorrect terminology.

I have two ESXi servers. One of them is intended to be a more "traditional" setup, with lots of VMs running simultaneously and ESXi distributing resources automatically to the different ones. This one I'm having no problems with. The other one, however, is intended to be a system where we essentially only run one VM at a time. I have four or five VMs which I want to be able to run individually, each of which is a virtualized version of a physical PC which requires a significant amount of processing power. The server we purchased for this one is a DL380 Gen9 with 16 Intel Xeon E5-2667 v4 CPUs @ 3.20 GHz and 48 GB of RAM. We got the processors we did because we wanted to match the "max case" processing requirements of the physical PC running our most complex project.

The software in question is a real-time operator training simulator. The need is for this simulator to be able to run in real-time with spare capacity to be able to go at least 2X real time. Additionally, we can also just tell the software to just run "as fast as possible," in which case it will use as much processing power is available to run as fast as it can. For the Windows 7 VM I'm testing right now, on the physical PC version of the same thing I'm able to get over 2.5X real time before maxing out the processing power of the PC. In the VM, however, the software is maxing out at just below real time. The total processor usage according to the VM bounces around between 80-90%. Meanwhile, the physical CPUs in the ESXi host are barely breaking a sweat. The "busiest" CPU according to the ESXi web client is occasionally going above 50%, with most of the others well below that. I know that if the VM were given more processing power, the software would immediately try to use it and run the simulator faster, but for some reason it's not happening.

What I'm trying to understand is when I'm running just one single VM, what is it that would make ESXi "decide" that the running VM does in fact require more processing power to be given to it? I'm not sure at this point if this is an issue I can resolve within ESXi, in the VM, or with the specific software I'm using within the VM. Thanks in advance for any help anybody might be able to give me on this, and I'm happy to provide more details if it will help!

[Gidrakos]

1) Does the VM guest OS think it has all the processing power, or is it only being presented with some portion of what the host has? The VM probably thinks it's running at 80-90% because it's only been given a small percentage of the back-end server.

2) Try going into the VM's settings and adjusting the CPU Reservation - this tells the host "No, really, give the VM at least this much at all times." See if the guest OS reports a different CPU amount after doing this.

3) You also can't forget that the overlaying hypervisor running on the Host is going to take a little bit of CPU, and that slightly decreases performance when compared to running something on bare hardware.

[KMarn]

1) Yeah, the issue is that the VM is maxing out the resources it's been given by ESXi, so in Task Manager the CPU performance is showing consistently 80-90% usage. However, there is still plenty of physical processing power available in the ESXi host. According to ESXi, the VM is only using ~6 GHz out of a total available ~50 GHz when I have no other VMs running. What I'm not clear on is whether the VM somehow should be "asking" for more processing power, or if ESXi should be automatically providing more because the VM is using almost all of the processing power that's been given to it.

2) I tried this and it had no effect. After looking into it online, the Reservation setting has no effect on anything unless the physical host's CPU is maxed out by multiple VMs. For one single VM it doesn't actually do anything. I was hoping this would be a way to manually force ESXi to provide more processing power, but it wasn't to be.

3) I'm fine with some amount of overhead, but right now the software on my VM is stuck running about 40% of the max speed it can run on the physical PC. Plus the ESXi host is showing a ton of spare processing capacity, so I don't think ESXi's overhead is my limiting factor here.

Thank you for the suggestions!

[KMarn · Accepted solution]

I found my problem! The issue seems to be with the software I'm running on the VM. It seems to have some limit to only use most of a processor without truly maxing it out. The ready percentage in ESXi was almost always 0%, so ESXi simply didn't see a need to provide more processing power. I got some software for stress-testing a CPU, and as soon as I ran it, ESXi immediately started providing maximum CPU resources to my VM. So this isn't an issue with my configuration of ESXi or the VM at all.

-----------------------------------------

Was it helpful? Let us know by completing this short survey here.