Using a the Xen Hypervisor to Supercharge OS Deployment

Hypervisor technology presents some promising opportunities for optimizing Linux deployment. We discuss modifying initrd and the init process to work in cooperation with the Xen hypervisor to create any number of unique server instances by re-using a single system image. By isolating a server's unique properties into a set of patches to initialization scripts and other selected files, deployment of a new server can occur in a few seconds. To capture changes to a server's configuration that occur while it is running, we discuss using an LVM snapshot to hold changes to selected files. By separating the initialization and file data that make a linux server instance unique, that data can be stored and retrieved in a number of ways. We discuss storing and retrieving different initialization patches over the network and integration of these capabilities into the Xen tools. Potential uses for the techniques demonstrated in the paper include capacity on demand, and new methods of provisioning servers and workstations.