Live Migration: SSH bottleneck

[blackline]

I have three PVE in a cluster and would like to utilize the 50Gbit network between them for live migration. I only see about 400-500 Mbps, where it should be up to 6000 Mbps. Running top I saw that the ssh process was stuck at 100%.

Is it possible to use nc or some other, more performant method to transfer the data?

 

[_gabriel]

what physical cpu model ?

 

[fabian]

if you trust your local network, you can set the migration network and mode to use plain TCP (or setup something like wireguard, and set the migration network accordingly to tunnel plain TCP over that).

 

[spirit]

ssh tunnel is single threaded, AFAIK, it should be around 2~3gbits on modern cpu

edit /etc/pve/datacenter.cfg and set "migration: type=insecure" to disable ssh tunnel.

You should easily each 10gbit/s

 

[blackline]

More than that Thanks, this is amazing!