I have VirtualBox 6.1.4 running on macOS Mojave. I'm working with a Windows 10 guest and it seems I have some performance problems with disks.
Testing with Aja System Test, on macOS I see numbers around 2GB/s for both reading and writing (SSD with APFS on an unencrypted volume). Using the same software for measuring performance in the Windows 10 guest I see numbers such as (W/R):
50/70 180/60 170/70 70/58 88/53
The measurements have been taken in a row, each one takes only a few seconds. Of course I took care that no other process was running both in the host and the guest, so the first strangeness is the great variance in numbers in such a short time. Second, the read speed is always smaller than write speed; third, sure I don't expect to see the native performance, but I'm seeing a 10x reduction in the best case (which seldom happens) and 40x in the worst. I guess/hope this could be better (and consistently better).
Some note about the context. I do have a long experience with VirtualBox, but almost exclusively with Linux guests. A few years ago I worked with Windows 7 and 8, which were very different. I do have a Windows 10 guest that I seldom use, but not for high performance jobs. Now I got a VM made by others which I need to use for development and the disk sometimes acts as a bottleneck.
The VM I received was configured for use with VMWare: I deactivated all the relevant services and startup executables. I had troubles with the installation of some Windows updates (which I didn't want, but they managed to be applied): they caused strange problems, but they seem to be gone. GAs have been reinstalled after each update.
Host hardware:
- MacBook Pro (Retina, 15-inch, Mid 2015),
- 2,8 GHz Intel Core i7 (4x2 cores),
- 16 GB 1600 MHz DDR3;
- almost 200 GB free on the SSD.
Guest config:
- 8MB RAM, 3 processors,
- VT-x/AMD-V, Nested Paging, Hyper-V Parav.
- SATA Port 0: 60GB VDI
- SATA Port 1: 14GB VDI
I seem to have another problem, but I don't know whether it's related (at the moment I don't think): the CPU load on the host often seems disproportionate with respect to the CPU load seen on the guest. Anyway when I run the disk performance test CPU on the host is around 50/60%, but the system is not CPU-bound.