When my Intel S1200 server motherboard gave out, I figured a cheap replacement would get my humble ESXi server up and running again. An old Gigabyte GA-H97M-D3H I purchased second-hand gave my old Xeon CPU another home, and I couldn’t justify upgrading hardware to keep a homelab hypervisor running.
As it turned out, the hardware was actually the easy part. I’d stopped at ESXi 7 Update 2 after warnings from VMware about dropping support for Haswell-generation CPUs in versions 8 and 9. I’d since lost the installer for ESXi 7, and Broadcom, having acquired VMware in 2023, has made the public download unavailable.
So I installed Proxmox. It got the server running again, but would an ESXi guy like me feel at home using it?
I had to stop looking for the ESXi version of everything
The jobs were familiar, even when the names weren’t
I built my server with the working components from the old one. A Xeon E3-1225 v3, 16 GB of RAM, two small Intel SSDs, and a 1 TB mechanical drive. It’s nothing extravagant, but it’s enough to learn where Proxmox keeps things I’d normally find in ESXi without thinking.
Storage was the first big adjustment. I’d been so used to browsing datastores in ESXi, but Proxmox presented storage entries with different types and permitted content:
-
localholds files such as ISOs. -
local-lvmprovides thin-provisioned space for guest disks. -
directoryis file-level storage mounted on a standard Linux filesystem.
The installation SSD was assigned local-LVM (local), the second SSD to local-lvm (vm-ssd), and the mechanical hard drive was formatted as directory (bulk-hdd).
Networking looked a little more familiar with Proxmox’s Linux bridges acting like virtual switches that connected guest network adapters and physical interfaces. My vmbr0, assigned to NIC1, connects guests to my main 192.168.1.x LAN, while vmbr1, assigned to NIC0 is for my lab network 100.50.1.x.
While getting used to all the differences, I created a sort of cheat sheet to map out comparable roles:
|
What I knew from VMware |
What I found in Proxmox |
|
ESXi host |
Proxmox node |
|
ESXi Host Client |
Proxmox web interface |
|
Datastore |
Storage entity with a backend |
|
Standard vSwitch |
Linux bridge such as vmbr0 |
|
Physical uplink, such as vmnic0 |
Physical network interface assigned to a bridge |
|
VMware Tools / open-vm-tools |
QEMU Guest Agent for management, plus SPICE tools for desktop integration |
|
VMware Remote Console (VMRC) |
SPICE Remote Viewer |
Once I’d connected those terms to familiar ESXi jobs, I ended up spending far less time menu hunting, and more time getting on with configuring the server.
I brought a working VMware machine with me
The disk was imported, but the network needed its old identity back
I already have one good migration candidate in VMware Workstation. I’d previously rescued a physical mini PC by cloning it into an Ubuntu VM, complete with its Docker applications and a Caddy reverse proxy.
With the VM powered off, I created a full clone of its current state. That meant the snapshot chain was consolidated into a standalone VMDK. After transferring it to Proxmox, I checked the disk before importing it:
qemu-img info /mnt/pve/bulk-hdd/migration/mini-pc/mini-pc-p2v-cl1.vmdk
The disk had a virtual capacity of 128 GiB, but it only occupied about 19 GiB. I created a UEFI VM with the next available ID (102), then imported the disk into my SSD pool:
qm disk import 102 /mnt/pve/bulk-hdd/migration/mini-pc/mini-pc-p2v-cl1.vmdk vm-ssd
Ubuntu booted, but the n8n automations weren’t running at their familiar IP address. Ubuntu’s networking tool, Netplan, had a static configuration matching the old adapter’s MAC address. Proxmox had generated a different one, and that meant the guest had received a different IP from the DHCP server.
Unfortunately, Caddy still expected the original. Rather than reconfigure Caddy, I decided to just give the VM its old MAC address back:
qm set 102 --net0 virtio=00:0C:29:4A:E3:B4,bridge=vmbr0
After starting the Proxmox guest, its existing network configuration was applied and the n8n service loaded. My workflow was still there without needing to rebuild anything.
LXC gave my server a useful little side job
A browser-based LAN speed test didn’t need another full VM
Proxmox LXC gave me something different to try that doesn’t have a native equivalent on ESXi. LXC is a container that shares the host’s Linux kernel while providing its own Linux environment. Unlike the Docker applications inside my Ubuntu VM, this Debian container runs directly under Proxmox without needing a separate guest kernel.
I gave it one CPU core, 512 MB of RAM, and a 4 GB root disk, then gave it a useful test job. OpenSpeedTest would let me check local network speeds from a browser.
Inside the container, I installed the packages needed to support it:
apt install nginx git ca-certificates
Then I downloaded the OpenSpeedTest files:
git clone --depth 1 https://github.com/openspeedtest/Speed-Test.git /var/www/openspeedtest
I pointed Nginx at that directory and configured its upload handling. Opening the container’s IP address in my browser brought up the speed test.
With my laptop connected by Ethernet through a switch, it reported roughly 983 Mbps download and 980 Mbps upload. Of course, these were my local network speeds and, unfortunately, not my WAN speed.
It also got me thinking about other small jobs I could give these containers, like a local DNS server, dashboard, or even a Pi-hole project.
I trusted recovery after I actually tested it
A restored service meant more than a successful backup message
Once the mini OpenSpeedTest worked, I deliberately broke it by replacing its home page. I took a snapshot, replaced the page with a “recovery test” message, and refreshed the browser to confirm it had actually broken.
Then I shut down the container, rolled it back, and started it again. The speed-test interface returned, much like it would have when restoring a VM snapshot in ESXi.
That proved I could undo an experiment (provided I make a snapshot first), but I also wanted to test recovery from a backup archive. I backed up the container to my bulk-hdd using stop mode, which briefly shuts down the VM while Proxmox copies its data. By the end of the process, the resulting archive was 292 MB, and the log reported that the VM had started and was back online after ten seconds.
For the restore, I left the original container switched off and recovered the archive as a separate container with a new ID (104). It came up at the same IP address, and OpenSpeedTest worked happily again.
Snapshots aren’t a new concept for me after working with ESXi for years. But it was important to prove that I could recover a test guest in an unfamiliar system before I was faced with doing it for real.
A snapshot can undo an experiment, but it isn’t a great independent backup. My archive was stored on a separate disk in the same server, so even then I’d still need an off-host copy to protect against losing the entire VM.
I wanted my Linux VM to feel like VMware again
Clipboard sharing was only half the battle
I wanted to make sure I could replicate the ESXi VMRC (VMware Remote Console) clipboard sharing and desktop resizing experience with a Linux guest in Proxmox. I spun up a fresh Linux CachyOS guest VM, assigned its resources, and attached the ISO, then installed it.
Getting the desktop to behave like the VMware consoles I was used to took me a considerable amount of patience.
The QEMU Guest Agent handles management tasks and reports information to Proxmox similar to how open-vm-tools (VMware Tools) does in ESXi. Desktop conveniences such as cross-device clipboard sharing come from SPICE integration. Installing one, unfortunately, didn’t mean I had the other.
I first got text copying working through noVNC’s clipboard panel, but that’s clunky and awkward, and I wanted direct copy and paste. Switching to SPICE Remote Viewer and using Plasma’s X11 session gave me the clipboard behavior I wanted. Desktop resizing, though, remained stubbornly uncooperative.
The CachyOS guest could see the resolution being requested by the viewer, but wasn’t applying it. This command gave it a good nudge:
xrandr --output Virtual-1 --auto
Afterward, the desktop followed the window size, up until I rebooted. My eventual workaround was a login script that checked for a resize request every two seconds and applied the preferred resolution whenever it differed from the current one.
That was irritatingly fiddly and an annoying amount of work for something that just worked on VMware with VMware Tools. However, my Windows 11 VM supplied a bit of a reality check. Once configured for SPICE, clipboard sharing and resizing just worked, even after a reboot.
So, the awkward part was this particular CachyOS and KDE Plasma desktop. It definitely wouldn’t be fair to assume I’d get this experience with every Linux guest.
I came for a replacement and found a reason to stay
This old Xeon still does the jobs I need
This is only a small secondary homelab, and Proxmox VE has given me more than enough reasons to keep it busy. My existing hardware runs a migrated workload, a useful little container with the promise of more to come, and fresh desktop VMs. I’ve also tested getting a service back when something breaks.
I still miss the familiarity of VMware, and trying to get a Linux desktop to resize the way it does in VMRC took up a bit more time than I was prepared for. But that frustration hasn’t outweighed the fact that Proxmox has given me my server back.
I installed Proxmox because I couldn’t get my old ESXi box back, but I’m keeping it because I found more things I want this server to do.