I'm wondering, caker, what your thinking is on backup systems. I figure that the main problem is that you can't guarantee that the filesystem is in a coherent state, therefore simply making a binary copy of the running disk images doesn't work.

So is there going to be a special NFS mount or rsync server or something like that where you provide some Perl script that we have to run on a regular basis that needs to figure out which server to connect to, or is there going to be some sort of patch to UML that will snapshot the system?

Having complete binary copies is probably the most useful in the case of a total blow-out, but there are a few obstacles. One is what you mentioned, the Linode would be in a running state. That can be avoided by pausing the Linode, calling sysrq s (sync), and making the duplicate while the Linode is paused. I also lose the "sparseness" of the disk images once they're transferred to another machine. So, I'd need backup space for all the Linodes * the size of their disk images. (edit: scratch that, rsync has a -S --sparse option!)


I really don't have an answer for you yet. It's still something I need to think about

-Chris

At another VDS provider I know, they supply backup space if you request it, but leave it up to you what to copy to it and when, etc. The space you get is equal to your allocated space.

Might this be a good compromise?

I, as another user, wouldn't mind that.

I mean, Caker's got a whole racket built up here around a bunch of folks who want to support themselves, more-or-less, so letting us do our own backup, especially if it is built around one of the usual tools for slinging files around (scp, rsync, nfs, etc)

I would like this option. I wouldn't need a backup of all of my disk images, and I wouldn't need a backup that often because most of my data doesn't change.

caker - do you have any thoughts on this? I'm curious to know how you feel about it.