Text version of the portfolio of Rémi Moreau, a BTS SIO SISR student (French two-year degree in IT infrastructure and networks). Open in RémiOS, the interactive version → · Version française

Troubleshooting the Proxmox servers in the lab room

Type
School project
Date
4 November 2026
Status
completed
Technologies
Cisco IOS, Proxmox, Debian, Wireshark, PowerShell, netsh, Windows, Claude
Skills
B1.1 Manage IT assets; B1.2 Respond to incidents and to requests for support and change; B1.5 Make an IT service available to users

Context

A BTS SIO practical session. For this lab, we needed the servers installed in the room's rack: one per POD (POD01, POD02 and POD03), each connected to its own Cisco switch. The servers had just been set up, but they were not working: connected to no network, without a screen, and nobody knew their IP address. There was no way to connect to them and start the lab.

As I was ahead on the lab, I offered to take care of it. With the teacher's agreement, I tried to fix the problem on POD02, including work on the switch, then prepared what was needed to repeat the procedure on the other two PODs.

Objectives

Documentation to download

The documentation also serves as a procedure if the problem comes back: it starts with a quick procedure and a "symptom, cause, what to do" table, then tells the troubleshooting step by step, and ends with a glossary and a summary of the commands.

The hardware

Item Model Role
Switch Cisco Catalyst 2960-XR, IOS 15.2(5b)E Connects the POD's machines
Server Dell Precision Tower 5810, Proxmox VE The lab's hypervisor
Workstations MacBook, Windows laptop Administration workstations, with a USB-C to RJ45 adapter and a USB-serial console cable

The POD network is isolated: no internet box or DHCP server, so no address is assigned automatically.

POD02 rack: Cisco Catalyst 2960-XR switch at the top, Cisco 1921 router below, blue cable from the workstation plugged into the switch
The POD02 rack: Catalyst 2960-XR switch and Cisco 1921 router

Implementation

The principle. Without DHCP, a workstation can only talk to the server if it is in the same subnet, with a static IP address. So I had to plug a workstation into the switch and, first of all, find the server's IP address.

The switch, through the console. No port LED lit up and the SYST LED was blinking green: according to Cisco's documentation, the switch was still running its power-on self-test (POST). Through the console cable (screen at 9600 baud), I saw that it was stuck at the bootloader's switch: prompt, without having loaded its operating system. The boot command loaded IOS and the switch started. With show interfaces status (after terminal length 0, which turns off paging): no port disabled, all in VLAN 1, the server on Gi1/0/14 and the workstation on Gi1/0/16.

Front panel of the Cisco Catalyst 2960-XR switch: SYST LED, MODE button, MGMT and RJ45 console ports, mini-USB console port
The switch front panel: SYST LED, MGMT port and console ports

PortFast. Each time a cable is plugged in, the STP protocol blocks the port for about 30 seconds, while it checks that there is no network loop. I enabled PortFast on the server and workstation ports (spanning-tree portfast), so that they forward traffic straight away, then saved the configuration with write memory. IOS stores it as spanning-tree portfast edge: edge means a port connected to a machine, never to another switch.

Finding the server. Wireshark showed no ARP request from the server, which would have given me its IP. On the switch, show mac address-table interface gi1/0/14 gave its MAC address (a Dell prefix), and its counters showed only 34 packets received: a server that has no one to reach stays almost silent.

The solution: IPv6 link-local. On Linux, every network card has an IPv6 "link-local" address (fe80::…), even when nothing is configured, and Debian, which Proxmox is based on, derives it from the MAC address (EUI-64 method). For example, a fictitious MAC 00:11:22:33:44:55 gives fe80::211:22ff:fe33:4455. I worked out the server's address the same way, then connected over SSH from the Windows PC: ssh root@fe80::…%12, where 12 is the number of the PC's Ethernet card. The ip a command gave the server's IPv4 address, a /16, carried by vmbr0, Proxmox's virtual switch.

The school network's IP and MAC addresses are hidden in this page, in the photos and in the downloadable documents.

SSH terminal connected to the POD02 server: ip a shows the IPv4 address of vmbr0 and the link-local IPv6 address, blurred
Connected to the server over SSH: ip a gives its address (blurred)

For POD01 and POD03. A PowerShell script, Get-InfosPod.ps1 (available for download above), repeats the whole procedure from the Windows PC: it finds the Ethernet card, discovers the server over IPv6, reads its configuration over SSH, queries the switch through the console (-Cisco option), puts the workstation in the right subnet (-ConfigurerPC option) and writes a report.

Troubleshooting carried out with the help of Claude (AI), which also wrote the script and formatted the documentation. My role: working on the hardware, making the choices and testing.

Results and testing

Difficulties encountered

An Ethernet adapter the Mac could not see. The Mac did not detect the USB-C to RJ45 adapter: the "Ethernet Adapter" entries listed by networksetup -listallhardwareports were Apple Silicon virtual interfaces. On Windows, the same kind of adapter is recognised without a driver ("Realtek USB FE Family Controller"): I carried on with a Windows PC. The Get-NetAdapter command also helped me spot the right card among the virtual ones (VirtualBox, VPN).

PowerShell: Get-NetAdapter lists Ethernet 5 (Realtek USB FE Family Controller, Up, 100 Mbps) and Ethernet 6 (VirtualBox Host-Only Ethernet Adapter)
Spotting the real Ethernet card among the virtual ones

Wireshark filters rejected. dhcp and udp.port == 67 turned red: I was typing them into the capture filter, which does not use the same syntax as the display filter (arp or udp port 67 on one side, arp || dhcp on the other). Solution: capture without a filter, then filter the display.

"Invalid input" on the switch. The PortFast configuration was rejected: I was in user mode (Switch>), which only allows viewing. You first have to switch to privileged mode with enable (Switch#), then configure terminal.

A server too quiet for Wireshark. With no traffic to capture, there was no way to read its IP from ARP requests. That is what led me to the IPv6 link-local address, derived from the MAC address read on the switch.

An open question. Why does the switch stop at the bootloader? show boot rules out a manual boot setting (Manual Boot : no), and the path to the IOS image is set. However, Enable Break : yes: a "Break" signal received on the console during startup, for example when a console cable is plugged in, can be enough to interrupt it. Next step, with the teacher's agreement: check that the image exists (dir flash:), disable this option (no boot enable-break), then test a restart (reload).

Skills applied