I’ve been trying out Xen for my new colo machine. It has a few rough edges, but it seems much nicer to use than UML. I’m planning to run an internal NFS service as one of the guest domUs, something I would never considered with UML. However, since a domU can get direct access to lvm block devices its pretty much as quick as running from the ‘host’ dom0.
Note, the host instance dom0 itself is a ‘guest’ running on top of the hypervisor.
An example of the performance difference between domU and dom0:
Timing cached reads: 2860 MB in 2.00 seconds = 1430.00 MB/sec
Timing buffered disk reads: 96 MB in 3.04 seconds = 31.58 MB/sec
Timing cached reads: 2200 MB in 2.00 seconds = 1100.00 MB/sec
Timing buffered disk reads: 94 MB in 3.02 seconds = 31.13 MB/sec
One of the things I’m thinking would be very good for Xen at the moment is more documentation.