Published on
Takes approximately 1 minute minutes to read

Cloning Disks across a Network

I was migrating one dedicated server to dedicated another in a different data center. I immediately reached for Clonezilla, as my general "disk cloner". Turns out that my data center provider limit their KVM network to ~400KiB/s, so actually booting Clonezilla into RAM look over an hour, for each server...

After using the standard remote_source/remote_dest options, I discovered that Clonezilla reads complete junk from my disks (root on btrfs RAID + luks) for some completely unknown reason (even with block copying enabled). I assumed it was from my somewhat exotic disk setup. Either way, that completely nixed Clonezilla from my list of options.

Reaching into my toolbox left me with dd. I've never dd over a network, but it seems rather trivial. In theory, I'll just bit bang blocks from one server to another.

So I booted both systems into a PXE rescue RAM disk and installed tmux as my screen tool. Ultimately, the command I used (on the target machine) was:

ssh root@host dd if=/dev/sdb bs=16M | dd of=/dev/sdb bs=16M status=progress

My disks were encrypted, so I figured that compression wouldn't be the most useful thing to do - I suspected I might be hitting SSH crypto limits alone (which is single threaded, if I remember correctly).

The command saturates my 1Gbps link at around 80%. Running two in parallel appears to make the transfer network bound.