Win7 VM performance very bad after PVE upgrade from 5.4 to 6.2

[mdo]

We upgraded PVE on older hardware from version 5.4 to latest 6.2-11 (latest) as in-place upgrade 2 days ago.

There are only 2 active VMs on that server currently, a Linux VM which seems to "behave" (perform) as before and an old Win7 VM (hopefully to be decommissioned within the next 4 week).

The nightly vzdump/snapshot for its 300GB virtual disk (ending in about 160GB .lzo file) was still running the next morning (after 8 hours) and we had to cancel this. It normally takes (took) less than 1 hour so something is extremely slow after this PVE (Debian) upgrade (?)

It's a very simple config (we believe):

qm config 200
boot: c
bootdisk: virtio0
cores: 2
description: Windows 7 Professional%0AHosting SQL Express for V6 Database
ide2: none,media=cdrom
memory: 8192
name: nlc005sv
net0: e1000=46:AE:42:A1:ED:26,bridge=vmbr0
onboot: 1
ostype: win7
sockets: 2
startup: order=3,up=20
virtio0: local:200/vm-200-disk-1.raw,format=raw,size=300G

 

[t.lamprecht]

Hi!

So it's rather the backup job performance which degraded, not the internal VM performance?
Or did that also suffer?

How do you backup, live?

A general thing which got my attention is that you do not use the guest agent, for a clean live backup that should help make everything more consistent - especially if this is running a SQL database.

BTW: As you plan to decommission this one it may not be worth the hassle, but if you have other Windows VMs around I'd suggest switching over to VirtIO SCSI for the disk (vs. your currently used, a bit slower and with less features, Virtio-Block), and VirtIO for the network. Check out: https://pve.proxmox.com/wiki/Windows_VirtIO_Drivers#Installation

 

[mdo]

Hi Thomas

thanks for your reply. It looks like impact on that VM also while running, not only the backup.

We follow all your recommendations like VirtIO SCSI and VirtiIO for NICs plus use the guest agent for all Windows VM installations in recent times. That old Win7 VM goes back to some old, PVE starting times for us and I did not really want to touch it or make changes. Anyway, we added the agent but that didn't change the backup performance.

The local storage is on an Adaptec RAID 1 array (Sata disks) plus we have a single Sata disk attached to the mainboard that is normally used only temporarily during nightly backup operations. Single drive = non critical data normally.

I have temporarily allowed this single disk drive to be use for disk images and restored the VM onto that drive and performance seems back to normal, a vzdump (snapshot) backup now taking about 1+ hr rather 6hrs the night before.

I suspect that the in-place upgrade affected the Debian drivers for that Adaptec RAID controller access - but don't know how to verify and fix this. We cannot leave the virtual disk on the single drive so have to migrate this back into the RAID sooner rather later. If I cannot find a solution, I will have to try a fresh PVE 6.x installation, re-allocate the RAID storage and restore our system customisations.

Regards,
Michael