[libvirt-users] High i/o-wait in guest but no i/o-wait on host

Hi, I have an issue with one of our CentOS 7 hypervisors. The system is a Dell server equipped with Samsung 840 Pro SSDs. What happens is that the guest does about 40MB/s of writes (mostly MySQL inserts) and ends up becoming almost unusable because of high i/o-wait numbers yet when I check on the host vmstat consistently shows i/o-wait being 0 the whole time. This is weird for two reasons: a) I would not expect 40MB/s to create such an extreme congestion on the SSDs. b) If that congestion is in fact real I would expect to see non-zero i/o-wait numbers on the host created by the corresponding qemu process. Right now it looks like the i/o requests get stuck in the guest even though there is no congestion on the host. Could this be a virtio bug? Has anyone an explanation for this strange behavior? Regards, Dennis