Proxmox low disk performance for Windows server 2016 VMs

[fusrohdan]

Hi all,

I am trying Proxmox for the second time and having the same issue (So probably my fault this is happening) I have a "server" with the following specs:

I5 2500K (Not overclocked), 32GB DDR3, 1 500GB ssd for Proxmox and ISOs, 2 500GB SSD in ZFS raid , 2 1TB Baracuda compute hdd in ZFS Raid 1, 1 4TB decent-ish disk.I installed Proxmox on this today using the latest ISO.

So I have multiple VMs running. Mostly for training for my MCSA in server 2016. When I had this host running Server 2016, read/write speeds to the 4TB disk over the network were always about 80-90 Megabytes per sec. I have tried to mirror my previous setup so I have a VM with a single core, 1.4GB ram, 50GB disk on SSD and 3TB disk on 4TB. When I am copying onto the 4TB disk from my PC I am getting around 30 Megabytes per sec, but then the speed keeps diving to 0 and freezing. I am copying large files, mostly ISOs and films.

Here is the config for the VM:
Processor: 1, type host, AES instruction set on,
Memory: 1.46 GiB ballooning off
BIOS: OVMF
Display: Standard VGA
Machine: q35
SCSI controller: VirtIO SCSI (Drivers installed in guest)
Disk1: 50GB on raid 1 ssd, no form of caching (Drivers installed in guest)
Disk2: 3TB on 4TB disk, no form of caching (Drivers installed in guest)
Network device: virtio bridge, firewall=1 (Drivers installed in guest)



Any idea what is causing this? I tested the file transfer speeds to and from the old server to the same disk and the transfer speed were maxing the disk out, now on proxmox it is nowhere near.

 

[AllanM]

1 CPU and only 1.46GB RAM?

MS recommends an absolute minimum of 2GB with desktop experience installed.

I would give it 2 cores and 8GB RAM and retest before chasing this problem anywhere else.

Also.... with CPU type "host" selected for a VM, it should pass through ALL instruction capabilities of your host CPU without the need to manually set flags. No need to select "AES."

-----------------------

We used to have a domain controller running at work. The MSP who configured it only gave it 1 core and like 4GB RAM. It was often unresponsive/hung/crashed because it didn't have enough resources to work with.

--------------------

On a separate note.... Baracuda Compute drives... many of those are SMR drives. I realize this isn't applicable to this test (not the drive in play), but these will be poor performers on write operations. Something to keep in mind.

 

[fusrohdan]

Hi Allan,

Thank you for your reply and insight. I have taken your advice and added an extra core and upped the ram to 8GB.

The issue still persists. Although it does seem to be slightly better now. I was watching task manager inside the VM. The disk was writing at
111 mb/s for say 10 sec, then dropping down to 50 mb/s then 30 mb/s. At this point my pc says transfering at 0mb/s. When at this point, the disk activity both both the C(SSD) drive and D(4TB) sat at 100% but transfer rate according to task manager was 0kb/s. Also, I have reinstalled this VM before you replied, I set the controller to VirtIO SCSI single and enabled IOthread for each disk.

What confuses me the most about this issue is the fact when I have the same hardware running with Server 2016 as the Host, I never had this issue and performance was awesome, even with 1 core and low ram.

Interesting about the Baracuda drives, looks like some more research is needed

 

[AllanM]

Hi fusrohdan,

Have you run a disk performance test within the VM... like, a file copy test? I would get a few tens of GB worth of big files duplicated in a folder and do some copy operations on the windows server VM itself. Take the network out of the equation. Figure out whether the performance issue is related to network/smb/etc, or actual disk performance issues... This might be a fundamental difference in the way proxmox/zfs are handling writes. If ZFS/proxmox is forcing complete write-sync for every bit through the path with no caching, this can reduce write performance to kilobytes/sec on some drives.