Avi Kivity wrote:
Bill Davidsen wrote:
Based on a discussion in another list on ioschedulers, I added a few
lines to my rc.local to set the scheduler to "noop" for disks with
model QEMU*, to the scheduling in the VM will not be computing with
the scheduling in the host.
I wonder if the same thinking applies to the "rotational" feature
added to /sys/block in 2.6.29, intended for SSD but probably
appropriate for virtual and nbd devices as well.
A virtual disk backed by a rotational physical disk probably should be
marked rotational, at least with cache=off.
The question was more regarding virtual drives backed by files, but even
so I would think the bad case would be to let the host decide what
methods should be used, rather than the VM, which is more likely to do
things wrong based on incomplete information.
Feel free to clarify if you think there's a case where the VM and the
host should both be doing caching and scheduling. The only one I can
imagine is if a USB port is passed to the VM and it has a storage device
attached. That's not really a common configuration, I would guess.
--
Bill Davidsen <davidsen@xxxxxxx>
"Woe unto the statesman who makes war without a reason that will still
be valid when the war is over..." Otto von Bismark
--
To unsubscribe from this list: send the line "unsubscribe kvm" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
