WinServer 2022 (and Win11) slow performance on 5 nodes cluster

[prxmxEDS]

Guest VM (i.e. Windows Server 2022 with VirtIO driver and QEMU agent installed ver. 0.1.271)
looses performance migrating from old CPU type (hstp-pve01 and hst-pve02) to "less older" CPU type
(hst-pve04, hst-pve05, hst-pve06).
For example, executing PowerShell ISE took about 1.5 secs to open on first two hosts
while more than 2 secs on the other three hosts.
We tried to change different CPU types without any success.
Below info on hw & config

all 5x hosts are running (Boot Mode: Legacy BIOS)
Proxmox VE Virtual Environment 8.4.21
Linux 6.8.12-43-pve (2026-08-18T13:29Z)
pve-manager/8.4.21/2606ac850d46da29

same storage for all VM's:
each host has 2x SFP+ 10 Gbps (LACP, MTU 9000) connections
to a NFS server running on Debian 13.4.0 Trixie on a
DELL PowerEdge R730xd
2x socket Intel XEON E5-2690 v4 @2.60GHz
64GB RAM
PERC H730P Mini: RAID Controller
12x HDD DELL Enterprise 3,5 6TB SAS 6Gbps 512e RPM 7.2K
...RAID6 with /dev/sda4 partition for NFS type btrfs

...resources on 5x hosts
hst-pve01
DELL PowerEdge R720 (March 2014)
32 x Intel(R) Xeon(R) CPU E5-2640 v2 @ 2.00GHz (2 Sockets)
128GB RAM
CPU usage: 3%
RAM usage: 35%
VM's: 1x WinServer 2022, 2x Win 11
hst-pve02
DELL PowerEdge R720 (March 2014)
32 x Intel(R) Xeon(R) CPU E5-2640 v2 @ 2.00GHz (2 Sockets)
128GB RAM
CPU usage: 5%
RAM usage: 28%
VM's: 3x WinServer 2022, 1x Debian
hst-pve04
DELL PowerEdge R830 (October 2016)
80 x Intel(R) Xeon(R) CPU E5-4620 v4 @ 2.10GHz (4 Sockets)
512GB RAM
CPU usage: 4%
RAM usage: 15%
VM's: 1x WinServer 2022, 1x Debian, 4x Win 11
hst-pve05
DELL PowerEdge R830 (December 2016)
80 x Intel(R) Xeon(R) CPU E5-4620 v4 @ 2.10GHz (4 Sockets)
512GB RAM
CPU usage: 3%
RAM usage: 7%
VM's: 4x WinServer 2022, 1x Debian
hst-pve06
DELL PowerEdge R630 (December 2014)
48 x Intel(R) Xeon(R) CPU E5-2690 v3 @ 2.60GHz (2 Sockets)
128GB RAM
CPU usage: 2%
RAM usage: 30%
VM's: 2x WinServer 2022, 3x Debian


Proxmox .conf file for VM Windows Server 2022:
agent: 1
balloon: 0
boot: order=scsi0;ide2;net0
cores: 4
cpu: IvyBridge,flags=-md-clear;+pcid;+spec-ctrl;+ssbd;+aes
ide2: none,media=cdrom
machine: pc-q35-9.2+pve1
memory: 16384
meta: creation-qemu=7.2.0,ctime=1690467894
name: srv-eds-rdh02
net0: virtio=BC:24:11:11:BC:3D,bridge=vmbr0
numa: 0
ostype: win11
scsi0: nas02_ds01:117/vm-117-disk-0.qcow2,cache=writeback,discard=on,iothread=1,size=100G
scsi1: nas02_ds01:117/vm-117-disk-1.qcow2,cache=writeback,discard=on,iothread=1,size=300G
scsihw: virtio-scsi-single
smbios1: uuid=f0e0e55a-890b-4e5b-9c3a-de32978d3801
sockets: 1
tags: rds
vmgenid: f0b7ac57-f9a4-405b-8300-a1cdb3d945d2


Worst... some applications that run .exe on UNC link are
very very slow on Windows 11 and not usable by users even if the
VM is hosted by hstp-pve01 and hst-pve02.

Please, may someone clarify this?

Thanks

 

[ghusson]

Hello, could it be a disk read cache that is present on the source server but not on the destination server ?
If you want to clear caches, you can refer to this link for example : https://linuxvox.com/blog/linux-cache-flush/

 

[cwt]

You got a „chain of don’t-do-this“ in your config. Rotating disks on hardware RAID6 (double parity) -> BTRFS -> NFS connection -> and finally qcow2 for the VMs. That’s the biggest (mis)config in your setup. The change in CPUs (more cores but slower speed) doesn’t really matter.

 

[prxmxEDS]

Thnks Ghusson,
I've tested the datastore on all nodes using 100MB, 500MB and 1GB files:
#!/bin/bash

sync;sync
echo 3 > /proc/sys/vm/drop_caches
echo "All caches have been flushed."
echo -e '\n'
echo "writing file on local disk..."
dd if=/dev/zero of=./testfile-pve01_1GB bs=1M count=1000 oflag=direct
echo -e '\n'
echo "writing file on NFS volume..."
time dd if=/dev/zero of=/mnt/pve/nas02_ds01/clonezilla/visi/testfile-pve01_1GB bs=1M count=1000 oflag=direct

Attached you may see the results.

 

[prxmxEDS]

Thank you CWT,
what do you think is the better solution for datastore?
Of course we would like to keep the hardware we have!

I already thought to change the fs from btrfs to ext4 but I don't know if we will boost the disk sybsystem.

About the CPU: we notice difference performances with and without the md-clear flag especially on Win11 boot.
The logon screen took more (MORE) time to appear without -md-clear flag.

Regards