Windows Server 2019 Slower than other VMs

Badge +3

I am having the same issue on more than one host (Dell and HP), with more than one VM. Windows server 2019 seems much slower in Nutanix than for instance normal kvm in Debian (Proxmox) on the same hardware. I have been trying the same hardware and workload to compare Proxmox with Nutanix.  In every respect, every test, Nutanix is faster, but not when running 2019.  In some cases, other windows server version will run well (older the OS the better) Is this a driver issue ? Cpu features ? Any ideas about what to check ? In all cases I am running ce-2019.11.22-stable and VM’s have Nutanix-VirtIO-1.1.5-64bit

This topic has been closed for comments

7 replies

Userlevel 3
Badge +4

Slower performance? Slower boot, or more CPU/RAM spikes? Less IO? Did you use legacy boot or UEFI boot?

Badge +3

A valid question indeed, I did notice when I create a blank new VM using legacy for 2019 gave me issues. I prefer UEFI boot for 2019, but some legacy VM’s have bios setup (imported from Proxmox) So it seems to be on both ?

Userlevel 3
Badge +4

hmm, that's indeed a bit strange. So new VMs and imported are equally "slow”…?

Badge +3

Yes, on servers where the issue occurs, it seems to be both imported and newly created virtual machines that are affected. 

This issue appears only on some servers.

After more testing I found that it seems to be the hypervisor part that is slower. 
( I am comparing Nutanix to Proxmox, a debian kvm qemu hypervisor) 

I have found a “solution” though, one that gives me all the performance. 

--- WARNING --- 

What I am about to describe is an unsupported solution. 
I am doing this in a Lab as a demo. 
If you use this method you are on your own, as there is no support or documentation on this for CE ( or proxmox) 

(although for both the features exist) 


I got the best speed out of my server by installing Proxmox, and then nesting Nutanix in that. 
Nutanix storage fabric is used to hold the VM storage files. (drives are passed through directly to Nutanix) 
Proxmox is used as the hypervisor, accessing the VM store files over NFS. 

Doing this I can see that the Nutanix storage fabric is faster than conventional storage, so no slowness there. 
I can also see that the Proxmox hypervisor is (on my problem servers) running the VM’s better. 

I still don’t know why though. 

Badge +3

One thing I noticed, for the same workload, the actual cpu usage is less with the Proxmox hypervisor than with AHV (on the host) 

So the VM shows the same CPU usage on it’s task manager, but on the total usage of the host it seems AHV uses about 20% more CPU. 

Could AHV be doing CPU emulation ? 

Userlevel 5
Badge +4

AHV does cpu emulation,it’s full virtualization,  it has both software assisted and hardware assisted virtualization between Do you have SSD storage on your server or Did you emulate HDD to SSD?

Badge +3

For the goal of best build possible with old hardware (dell R720’s and HP Proliant DL’s ) I went for SSD’s, new, real datacenter grade SSD’s. 

The SSD’s and HDD’s get passed through as hardware to the nested Nutanix vm’s, not only for portability, but also flexibility (these drives could used for Nutanix “on metal”’ or as VM )

The other hypervisor can also do both type 1 or type 2 virtualisation for the cpu, and allows for easier selection of the CPU presented to the VM   

So  the thing that started this tread is difference between when the 2 different  hypervisors in terms of when they decide to (go level 1 or level 2) straight to kernel or emulate.  

I think the servers that are running slower when hosting in AHV have CPU’s of a certain age, and Nutanix is not designed to run on Xeon’s from yesteryear. ( 5 or more years old) so it ends up going for emulation rather than host kernel access. 

If I am not mistaken, Nutanix (the full deal, not CE) is first and foremost a storage fabric, and the AHV is just “extra’s” so you don’t NEED 3’rd party hypervisors, but there was a time there was no AHV, so the deployment of the combination of the 2  products for best total effect is not that crazy / foreign. 

But CE is supposed to be used with AHV, since development and testing is partly the motivation for having a Community Edition in the first place. 

However, I am building one kick ass demo, giving the client a taste of how their current workloads perform on Nutanix using old out of warranty hardware.