27 Setting Up a KVM VM Host Server #

virtio-scsi is an advanced storage stack for KVM. It replaces the former virtio-blk stack for SCSI devices pass-through. It has several advantages over virtio-blk:

qemu-system-x86_64 [...] \
-device virtio-scsi-pci,id=scsi

The vhost-net module is used to accelerate KVM's paravirtualized network drivers. It provides better latency and greater network throughput. Use the vhost-net driver by starting the guest with the following example command line:

qemu-system-x86_64 [...] \
-netdev tap,id=guest0,vhost=on,script=no \
-net nic,model=virtio,netdev=guest0,macaddr=00:16:35:AF:94:4B

Note that guest0 is an identification string of the vhost-driven device.

As the number of virtual CPUs increases in VM Guests, QEMU offers a way of improving the network performance using multiqueue. Multiqueue virtio-net scales the network performance by allowing VM Guest virtual CPUs to transfer packets in parallel. Multiqueue support is required on both the VM Host Server and VM Guest sides.

Tip: Performance Benefit

The multiqueue virtio-net solution is most beneficial in the following cases:

• Network traffic packets are large.

• VM Guest has many connections active at the same time, mainly between the guest systems, or between the guest and the host, or between the guest and an external system.

• The number of active queues is equal to the number of virtual CPUs in the VM Guest.

Note

While multiqueue virtio-net increases the total network throughput, it increases CPU consumption as it uses of the virtual CPU's power.

The resulting qemu-system-ARCH command line will look similar to the following example:

qemu-system-x86_64 [...] -netdev tap,id=guest0,queues=8,vhost=on \
-device virtio-net-pci,netdev=guest0,mq=on,vectors=10

Note that the id of the network device (guest0 ) needs to be identical for both options.

Inside the running VM Guest, specify the following command as root:

ethtool -L eth0 combined 8

Now the guest system networking uses the multiqueue support from the qemu-system-ARCH hypervisor.

Directly assigning a PCI device to a VM Guest (PCI pass-through) avoids performance issues caused by avoiding any emulation in performance-critical paths. VFIO replaces the traditional KVM PCI Pass-Through device assignment. A prerequisite for this feature is a VM Host Server configuration as described in Important: Requirements for VFIO and SR-IOV.

To be able to assign a PCI device via VFIO to a VM Guest, you need to find out which IOMMU Group it belongs to. The IOMMU (input/output memory management unit that connects a direct memory access-capable I/O bus to the main memory) API supports the notion of groups. A group is a set of devices that can be isolated from all other devices in the system. Groups are therefore the unit of ownership used by VFIO.

Important: No Hotplugging

As of SUSE Linux Enterprise Server 12 SP5 hotplugging of PCI devices passed to a VM Guest via VFIO is not supported.

You can find more detailed information on the VFIO driver in the /usr/src/linux/Documentation/vfio.txt file (package kernel-source needs to be installed).

VM Guests usually run in a separate computing space—they are provided their own memory range, dedicated CPUs, and file system space. The ability to share parts of the VM Host Server's file system makes the virtualization environment more flexible by simplifying mutual data exchange. Network file systems, such as CIFS and NFS, have been the traditional way of sharing directories. But as they are not specifically designed for virtualization purposes, they suffer from major performance and feature issues.

KVM introduces a new optimized method called VirtFS (sometimes called “file system pass-through”). VirtFS uses a paravirtual file system driver, which avoids converting the guest application file system operations into block device operations, and then again into host file system operations.

You typically use VirtFS for the following situations:

qemu-system-x86_64 [...] \
-fsdev local,id=exp11,path=/tmp/2,security_model=mapped3 \
-device virtio-9p-pci,fsdev=exp14,mount_tag=v_tmp5

sudo mount -t 9p -o trans=virtio v_tmp /mnt

Kernel Same Page Merging (KSM) is a Linux kernel feature that merges identical memory pages from multiple running processes into one memory region. Because KVM guests run as processes under Linux, KSM provides the memory overcommit feature to hypervisors for more efficient use of memory. Therefore, if you need to run multiple virtual machines on a host with limited memory, KSM may be helpful to you.

KSM stores its status information in the files under the /sys/kernel/mm/ksm directory:

tux > ls -1 /sys/kernel/mm/ksm
full_scans
merge_across_nodes
pages_shared
pages_sharing
pages_to_scan
pages_unshared
pages_volatile
run
sleep_millisecs

For more information on the meaning of the /sys/kernel/mm/ksm/* files, see /usr/src/linux/Documentation/vm/ksm.txt (package kernel-source).

To use KSM, do the following.