Thought it might be interesting to share some details of a cluster I helped build and demo for the Novell SLE 10 launch at LinuxWorld.
A couple of commenters to Jeff Jaffe's blog, pointed out a potential problem with running many virtual machines on one physical server, creates a much larger outage should the server fail; you lose all those VMs. We agree and designed a solution into SLES10.
With SLES10, you can cluster physical servers and failover VMs from one to another; using traditional cluster resources to manage each VM. SLES10 supports clustering of Xen VMs and the following slides illustrate the Linux World demo - a four node cluster sharing storage over iSCSI, and running two virtual machines as relocatable cluster resoures. The VM OS images are accessible to all nodes thanks to the Oracle cluster file system, and the cluster software monitors the virtual machines to enable local restart and failover between nodes.