Guide

Proxmox VE for Network Labs: A Free Home-Lab Server for CML, EVE-NG & GNS3

Most CCNA and CCNP learners start on a laptop, running a couple of virtual routers inside VMware Workstation or VirtualBox. That works until the topology grows: the third or fourth device eats your RAM, the fans spin up, and you can't leave the lab converging overnight because you need to close the lid and go to class. Proxmox VE fixes that by moving the lab off your laptop and onto a dedicated machine built to run virtual machines all day. This guide explains what Proxmox is, how it differs from the hypervisor on your desktop, and how it fits alongside the three tools most home labbers use: Cisco Modeling Labs (CML), EVE-NG, and GNS3. The goal isn't to sell you a server. It's to help you decide whether a Proxmox box is worth it for the kind of labbing you actually do.

Type-1 vs. Type-2: Where Proxmox Fits

Hypervisors come in two flavors, and the difference is the whole reason Proxmox exists as a category. A type-2 (hosted) hypervisor runs as an application on top of your everyday operating system: VirtualBox and VMware Workstation launch inside Windows, macOS, or Linux, and your VMs share whatever CPU and RAM your desktop OS isn't already using. That's convenient because you install it like any other program, but the OS underneath is always there competing for resources and keeping the machine busy with browser tabs, updates, and everything else.

A type-1 (bare-metal) hypervisor runs directly on the hardware with no general-purpose OS beneath it. Proxmox VE is exactly this: you install it onto a dedicated machine and it becomes the operating system. Under the hood it's a Debian-based system that uses KVM to run full virtual machines and LXC for lightweight Linux containers, so the Linux kernel itself acts as the hypervisor on the metal. There's no Windows desktop stealing cycles from your lab.

You administer the whole thing from a web UI in a browser pointed at the host's address, which means the server can sit headless in a closet with no monitor or keyboard attached. You reach it, and the lab VMs on it, over your home network from your laptop. Proxmox VE is free and open source; there's an optional paid support subscription for enterprises, but nothing about running a home lab requires it. Check Proxmox's site for current subscription tiers if you're curious.

Why Labbers Put Proxmox on a Dedicated Box

Always-on is the first win. A dedicated server lets you leave an OSPF area converging or a slow BGP lab running overnight and reconnect from any device in the morning. You're no longer tied to one laptop that has to stay awake and plugged in.

RAM headroom is the second, and it's usually the real bottleneck. Virtual routers and switches are memory-hungry; a single node can want anywhere from a few hundred megabytes to well over a gigabyte, and a full CCNP topology stacks many of them. A 16 GB laptop taps out after a handful of devices. A cheap used workstation or mini-PC with 32, 64, or more gigabytes runs topologies a laptop simply can't hold in memory, and CPU headroom keeps them responsive.

Snapshots make experimentation safe. You can capture a known-good VM state before you deliberately break routing, then roll back in seconds instead of rebuilding. Consolidation is the fourth benefit: on one machine you can run the CML appliance, an EVE-NG VM, a GNS3 backend, and a supporting Linux jump host or DHCP/DNS server side by side, each isolated in its own VM.

Cost seals it. The software is free and the hardware doesn't need to be new or fancy: an off-lease desktop or a small-form-factor mini-PC is plenty. Because everything runs on the server and you reach it over the network, your laptop stays light. It's just a browser and an SSH client, not the thing straining to hold the whole lab.

Running CML, EVE-NG, and GNS3 on Proxmox

The key thing to understand is that all three of these tools are virtual appliances, not desktop apps, which is precisely why a server that runs VMs is a natural home for them. Cisco Modeling Labs is distributed as an image you run in a hypervisor (an OVA plus a bare-metal ISO installer), not a program you double-click. Its nodes, such as IOSv and IOL, boot as nested VMs inside the appliance. The CML image runs on KVM, and hosting it on Proxmox is a well-trodden community path.

Here's the honest caveat: Cisco officially validates and supports CML on VMware, whether that's ESXi on bare metal or VMware Workstation, Fusion, or Player for personal use. Running CML on Proxmox/KVM works for many people but is community-supported, not a Cisco-blessed configuration, so treat it as such. VMware's personal-use desktop tiers are now free to download, which makes them the safe default if you want the supported platform, and VirtualBox is not a supported CML host. Before you commit hardware or troubleshoot a boot problem, read Cisco's current CML installation requirements, since supported versions and the personal-edition node limits change over time. And don't confuse CML with Packet Tracer: Packet Tracer is a free simulator that imitates device behavior, while CML runs real Cisco IOS images.

EVE-NG and GNS3 both ship as VM appliances designed to be imported into a hypervisor, so they drop onto Proxmox cleanly. EVE-NG is browser-based and fits the headless-server model perfectly. GNS3 uses a separate 'GNS3 VM' backend that you run on the server while the GNS3 GUI runs on your laptop and connects to it over the network, which again matches the offload-to-a-server pattern.

The one technical catch shared by all three is nested virtualization. Because these tools run VMs (the network nodes) inside a VM (the appliance), the Proxmox host has to expose its hardware virtualization to that appliance. That means enabling nested virtualization on the host CPU (Intel VT-x or AMD-V) and setting the appliance VM's CPU type to 'host' so the guest can actually see the virtualization extensions. Skip this and the router nodes either won't boot or will run unusably slowly. Essentially every modern Intel and AMD chip supports it; just confirm it's turned on.

Proxmox or a Laptop Hypervisor? Pick by Your Use Case

If you mostly spin up one or two routers for a quick CCNA lab and close it when you're done, a type-2 hypervisor on your laptop is the simpler choice. VMware Workstation or Fusion (free for personal use) or VirtualBox gives you one machine, nothing to rack, and nothing to leave running. Remember that VirtualBox can host EVE-NG and GNS3 appliances but is not a supported CML host, so reach for VMware if CML is your tool.

If you're building larger topologies, want the lab available around the clock, plan to run CML, EVE-NG, and GNS3 alongside each other, or you're moving into CCNP work where labs get big and long-running, a dedicated Proxmox server pays for itself quickly. It frees your laptop, hands you far more RAM and CPU than a portable machine can spare, and its snapshots and always-on nature match how serious study actually works.

Most people don't have to choose up front. It's common to start type-2 on the laptop and graduate to Proxmox the first time a topology stops fitting in memory. The skills carry over cleanly, because the appliances (CML, EVE-NG, GNS3) are identical either way. All you're changing is the machine that hosts them.

Frequently asked questions

Can I run Cisco Modeling Labs directly on Proxmox, or do I need VMware?

CML ships as a virtual appliance (an OVA/ISO you import into a hypervisor), not a native desktop install, so Proxmox can host it by importing the disk and running it as a VM. Cisco officially supports VMware Workstation/Fusion, now free for personal use, as the CML host and does not support VirtualBox; running it on Proxmox works for many labbers but sits outside Cisco's supported matrix. You must also enable nested virtualization so CML's reference node images can boot inside that VM.

Why won't my routers boot inside EVE-NG or GNS3 on Proxmox?

EVE-NG and GNS3 launch each network node as its own QEMU/KVM virtual machine, so the outer Proxmox VM has to expose the CPU's hardware virtualization (VT-x/AMD-V) to the guest, which is nested virtualization. Set the VM's CPU type to 'host' and enable nested KVM on the Proxmox node; otherwise the images fail to start or fall back to painfully slow pure emulation. This is the same wall that trips people up on other laptop hypervisors, not a Proxmox-specific bug.

How much RAM should a Proxmox lab box have for CCNA versus CCNP topologies?

Size by summing each node's requirement rather than counting devices: lightweight IOSv/IOL routers run in roughly 512 MB-1 GB each, while heavier images like CSR1000v/Cat8000v or IOSvL2 want closer to 2-4 GB. A small CCNA topology fits comfortably in 8-16 GB, but full CCNP labs with eight to a dozen nodes push you toward 32 GB or more. Check the current per-image minimums, since they shift between software releases.

Is Packet Tracer good enough, or do I actually need real images on a server?

Packet Tracer is a simulator that models device behavior rather than running real Cisco IOS, and it can auto-grade work inside pre-authored .pka activities, so it is fine for early CCNA fundamentals. But it omits or approximates many features, so CCNP-level topics need the real network OS images that CML, EVE-NG, or GNS3 run. That gap between a model and real code is the main reason labbers move to a dedicated Proxmox box.

Does running labs on Proxmox change how the devices themselves behave?

No. Proxmox only hosts the virtual machines; inside them you are running the same real network OS, so protocol defaults are unchanged. For example, HSRP preempt is still off by default while VRRP preempt is on, a next-hop static route still performs a recursive lookup while an interface-only route relies on proxy-ARP, and OSPF still advertises a loopback as a /32 — the hypervisor affects performance and management, not device logic.

Practice on real Cisco IOS

A fresh, graded Cisco Modeling Labs scenario every day — start with the free sample.