The cable was waiting for the disk: four VM round trips

October 24, 2026Real capture

✔ This is a real capture, made on October 24, 2026 with asciinema in a dedicated tmux session, with no editing. The migrations are real virsh migrate calls between stella and vega, on two test VMs with no operating system. The keystrokes were sent by a script and not typed by Ludo, the prompt is “ludo@labo”, only pauses longer than two seconds are shortened (23 minutes of tests become two and a half minutes), and the container ID in the tmux bar is replaced by a name of the same length. The terminal text and the narration are in French.

On the A400, toward stella, the cable and the switch end the same: the disk drops to 55 MiB/s from the second minute, once its SLC cache is full. Toward vega, the same model, rested, writes at 228 MiB/s. On the NVMe, the cable makes the trip in 42 s, the switch in 2 min 58 s, at gigabit speed. The terminal text is in French.asciinema-player ↗

The banc script sends a test VM back and forth between stella and vega, with virsh migrate --live --copy-storage-all, the same call vm-move makes. During each trip, it asks the receiving host every minute what its disk wrote and what its link received. The slower of the two sets the duration.

Four cases: the VM’s disk on the old SATA SSD (Kingston A400) or on the NVMe (WD SN730), and the data through the direct 5 Gb/s cable or through the 1 Gb/s switch. On NVMe, the network decides, and the cable wins by four. On the A400, the state of the receiving disk’s SLC cache decides.

The full explanation is in the article.