VMware vSphere: Pass-through WiFi and "This device cannot start" error (Code 10)

To conduct new research, I needed a WiFi card in a virtual machine. I have the WiFi adapter (Intel Dual Band Wireless-AC 8265) in my Intel NUC5i5MYHE, VMware has VMDirectPath I/O pass-through feature, sounds easy. But it was not easy.

After simple configuration of pass-through the WiFi adapter (including embedded Bluetooth device) appeared in Windows but didn't work. Device status was like this:

This device cannot start. (Code 10)

{Operation Failed}

The requested operation was unsuccessful.

At the same time embedded Bluetooth adapter works properly.

VMware vSphere: Does Hot Add and SCAv1/SCAv2 affect the performance of small VM?

I was wondering how the Hot Add and SCA features affect the performance of small VM (2 vCPU in my case). 

Enabling the Hot Add feature disables vNUMA feature that starts working when the number of virtual CPUs is greater than 8 and "can improve performance by facilitating guest operating system and application NUMA optimizations".

According to a good article "CPU Hot Add Performance in vSphere 6.7" this may reduce performance by 2 - 8%.

But does it affect VMs with a small number of vCPU? Let's check.

To mitigate “L1 Terminal Fault” vulnerabilities on Intel processors (CVE-2018-3646) VMware released KB55806. They propose two options:

1. ESXi Side-Channel-Aware Scheduler (SCAv1) for ESXi 5.5, 6.0, 6.5, and 6.7 prior to 6.7u2.
2. ESXi Side-Channel-Aware Scheduler v2 (SCAv2) for ESXi 6.7u2 (13006603) and later.

SCAv1 disables Hyper Threading, but SCAv2 does not. But all of them should affect performance, even though VMware says "Retain up to 100% performance for low to moderate loaded systems".

VMware vSphere: comparison of Paravirtual and LSI SCSI adapters

You may know that VMware vSphere offers several types of SCSI adapters for virtual machines depending on the OS. The main types are "LSI Logic SAS" and "VMware Paravirtual". There are a lot of articles on this topic (e.g. https://blogs.vmware.com/vsphere/2014/02/vscsi-controller-choose-performance.html). But the real comparison of the performance I saw once, but it's a bit outdated (ESXi 5.0 and Windows Server 2008R2).

Therefore, I did my tests (although not on the latest versions of OS) and compare IOPS, Latency and CPU usage. As a load generator used IO meter 1.1.0.