High Disk Latency in Windows Guest

[electricitymachine]

I'm running an HP ProLiant DL380 G9 with fully updated firmware and PVE 7, and for some reason, no matter what settings I use, I get a very high disk sec/transfer, which is an indicator of disk latency. The guest, Windows Server 2019, took half an hour to install in Proxmox, and took ages to load into the operating system. I tried installing PVE 6, ran apt update, no bueno. Still the same exact issue. I followed the wiki for Server 2019, tried various different cache options, and still the same result.

Win Server 2k19 performs perfectly when I install it bare metal. I am running a WD 4TB Red drive, no RAID, just a single disk. Virtualization is enabled in the BIOS. I'm at a complete loss. I'm not sure what commands to run or configs to post, but I am open to any and all suggestions. Thanks.

Bare metal:


Proxmox guest:

 

[Dunuin]

You should try virtio SCSI as your storage controller and SCSI as protocol if not already done yet. And you will need to install the virtio drivers from a second CD while installing Win or Win won't be able to use the virtual drives.
And you also might want to change CPU type from "kvm64" to "host" for better performance.
And you should check what kind of WD red you got. There are CMR and SMR drives. SMR aren't suitable as a VM storage because they get abyssimal latencies and writes as soon as the cache gets filled up.

 

[electricitymachine]

You should try virtio SCSI as your storage controller and SCSI as protocol if not already done yet. And you will need to install the virtio drivers from a second CD while installing Win or Win won't be able to use the virtual drives.
And you also might want to change CPU type from "kvm64" to "host" for better performance.
And you should check what kind of WD red you got. There are CMR and SMR drives. SMR aren't suitable as a VM storage because they get abyssimal latencies and writes as soon as the cache gets filled up.

Thanks for replying. It's an SMR drive, however, why would SMR be an issue if it's not even doing anything? That latency graph is just when it's idle, when it's not idle it peaks at 100 and stays above 50-60. VirtIO drivers are already installed, and I'm using VirtIO SCSI as the SCSI controller, the HDD is SCSI. Using host as the CPU doesn't change anything.

To better quantify how slow it actually is, I checked Resource Monitor, and the Disk Response Time ranges from 500 ms - 1500 ms, dropping down to 100 ms sometimes. When I did the same on bare metal, it was only 1-2 ms, and was much much faster.

Here's a screenshot of my hardware and options in the GUI:

 

[electricitymachine]

I did some more testing, and even with host CPU, it seems fine at idle, but even when doing a simple task such as browsing the Internet with IE, the graph has major spikes for avg disk queue length and sec/transfer:

The calm part is when I left it for 5 minutes to idle, and then I opened up IE.

 

[electricitymachine]

Update:

I haven't used Proxmox in a very long time. I went back to my machine, now equipped with an SSD, and installed a fresh copy of Windows 10. I compared LTSC with Windows Server, even tried Semi-annual Channel, but to no avail. The graphs look exactly the same. I even opened up Task Manager to look at the response time, and saw it peak at 1500ms for an SSD!

I'm at a complete loss now and haven't the slightest clue why this is happening. If anybody has any advice or needs me to run some command, I'll do anything to try and resolve it.

 

[fmielicki]

Hi, I just want to add that I'm facing the same issue on a Supermicro server. The virtual disk is on ZFS, guest is Windows 10. We're trying to run some ERP system with FirebirdSQL on it, and it's starting to become a problem.

 

[_gabriel]

edited : removed Cache advice.

smr or consumer ssd roughly works because they have small not protected write cache, which is not secure for data integrity, required in production environnement, in case of crash or power cut.