I really hate it when you dig down into some cool possibility, only to realize it isn’t possible. I’ve been playing a lot with Xen as of late. It’s really nice to be able to have another machine in a virtual sandbox. A while back I was talking with some guys from work about the possibility of using Xen in conjunction with AFS to create a very highly available compute cluster. Xen has the ability to migrate entire virtual machines from on physical machine’s memory to another physical machine’s memory, while the virtual machine continues to run and process stuff. It does not account for the “disk” associated with the virtual machine, which is where AFS comes in. AFS has the concept of cells, and the ability to move data on the server transparently to the clients. So it would seem that you could transfuse these two technologies to create virtual machines that would migrate around on various physical machines all transparently and automatically. But, here’s the snag. You can migrate Xen instances on the fly, but I don’t think there is currently any way to automatically fail over a virtual instance if a physical machines dies. And the same with AFS. You can migrate a R/W cell, and you can automatically fail over a R/O cell, but you cannot fail over a R/W instance. So, basically you can avoid downtime through scheduled maintenances, but can’t gain high availability through these technologies currently… :(