Dynamic writeback throttling

Sounds very nice. Has anyone had an eye on virtualized servers yet? If I run a SLES11 which runs 20 Linux VMs, it would be interesting to have some kind of global memory/disk pressure mechanism. If one out of the 20 VMs is under heavy pressure, a similar "pause others" would be welcome.

Also, some kind of auto-configuration for VMs would be nice, for example use the NOOP I/O scheduler, as the guest's can't optimize anyway, that must be done by the host and the RAID controller.

It seems to me developers still only/mostly look at single server or workstation performance, with a single user, and I wonder if VM behaviour couldn't be improved a lot.