I'm going to create a simple virtual hosts infrastructure using QEMU+KVM. It is expected to be very simple, without clouds or any other cool things. Right now the only requirement is the fault-tolerance for data (hosts' disk images online backup). Live migration welcomed, but is not obligatory. What way do you suggest? Right now I'm thinking about DRBD.
If the only thing you require from such a setup is having a replica on a second host, then DRBD is probably your best (and cheapest) bet.
Layers above depend mainly on what kind of setup you expect.
If manual failover is good enough for you, then setup drbd in master/slave mode and lvm, or pretty much any modern filesystem to store the images.
If you want some kind of automatic failover, or having some guests up on both machines (ie. for loadbalancing), you need some kind of clustered setup. Either cLVM, or clustered filesystem (such as GFS, or OCFS).
Another option, though I cannot recommend it yet, is using CEPH. It has not yet been declared production ready, and i don't expect it to be, until there is a recovery procedure for btrfs.