I migrated a node from Fremont to Dallas last night. As with pretty much everything at Linode, the process was extremely smooth, in fact it's amazing how easy it was.

The only issue I had was that it took quite a while to transfer the data. The maximum throughput rate was something like 1.8MB/s, which means it took around 2 hours to transfer my ~12GB node. Since I had to begin this procedure in the dead of night, it was kind of a pain.

Over a 100Mbit link, it should have taken something like half an hour. I believe the slowness is caused by SSH. Some of the buffers hardcoded into the default SSH cause serious bandwidth underutilization over high-speed, higher-than-LAN latency links.

Here's some research on this topic, including a patch: http://www.psc.edu/networking/projects/hpn-ssh/

I think it's at least worth looking into. Thanks!

Actually, we're already using those patches. With additional tweaks and customizations and those patches, we were able to improved same-DC migration speeds by 10x!

Unfortunately, inter-DC migrations go through the Internet, where we have no control over routes and link speeds. I'm afraid there's nothing more we can do.

-Chris

Great. I just wanted to make sure there wasn't anything being left on the table. Thanks for being on top of things.

Zero the empty space, pipe it through gzip.