15.1. Kernel-based Virtual Machine
15.1.1. Memory enhancements
- Transparent Hugepages increase the memory page size from 4 kilobytes to 2 megabytes. Transparent Hugepages provide significant performance advantages on systems with highly contended resources and large memory workloads. Additionally, Red Hat Enterprise Linux 6 provides support for utilizing Transparent Hugepages with KSM.
- Extended Page Table age bits enables a host to make smarter choices for swapping memory under memory pressure and allows swapping of Transparent Hugepages by breaking the extended pages into smaller pages.
15.1.2. Virtualized CPU features
- Red Hat Enterprise Linux 6 supports up to 64 virtualized CPUs for a single virtualized guest.
- CPU extensions present on the host processor can now be utilized by virtualized guests. Support for these instruction sets allow virtualized guests to take advantage of modern processor instruction sets and hardware features.
- The new
x2apicvirtual Advanced Programmable Interrupt Controller (APIC) improves virtualized
x86_64guest performance by allowing direct guest APIC access and removing the overhead of emulated access.
- New user space notifiers allow the caching of CPU registers, avoiding the computationally expensive actions of preserving register states of unused components during context switches.
- Read copy update (RCU) kernel locking is now uses enhanced symmetric multiprocessing support. RCU kernel locking provides greater performance for networking functions and multi-processing systems.
- The QEMU emulated block driver features support for fully asynchronous I/O,
pwritevfunctions. These functions increase performance for storage devices using the QEMU emulated block driver.
- The QEMU Monitor Protocol (QMP) allows applications to communicate with the QEMU Monitor correctly. QEMU provides a text-based format that can be easily parsed and support for asynchronous messages and capabilities negotiation.
- Indirect ring entries (spin locks) for the para-virtualized (virtio) driver improve block I/O performance and allows more concurrent I/O operations.
- Virtualized storage devices can now be added and removed (hot plugged) from guests during runtime.
- Support for block alignment storage topology awareness. Underlying storage hardware features and physical storage sector sizes (for example, 4KB sectors) are presented to guests. This feature requires compatible storage device information and commands. Guest topology awareness allows virtualized guests to optimize file system layouts and improved performance of applications using I/O optimizations.
- Performance enhancements for the qcow2 virtualized image format.
- The vhost-net feature moves various network functions from the QEMU user space into the kernel. vhost-net uses fewer context switches and vmexit calls. These enhancements improve performance of SR-IOV devices, directly assigned network device and other network devices.
- MSI-X support which increases the number of interrupts available to network devices. MSI-X support increases the performance of compatible hardware.
- Virtualized network devices can now be hot plugged and hot removed from running guests. Network boot using gpxe fora more advanced PXE network booting.
15.1.5. Kernel SamePage Merging
15.1.6. PCI passthrough
- Guest ABI stability provides enhanced migration support. Guests PCI device numbers are preserved during migration and identical PCI device positions are presented after migrating the guest.
- Migration now accounts for CPU models. CPU models allow guests to take advantage of new processor instruction sets. Guests can be migrated to hosts with a compatible CPU model.
- The vhost-net feature allows guests using SR-IOV to migrate to non-identical host configurations that also use SR-IOV devices.
- Enhancements to the migration protocol.
15.1.11. Guest device ABI stability
virt-v2vtool, enabling system administrators to convert and import virtual machines created on other systems such as Xen and VMware ESX.
virt-v2vprovides a migration path for Xen guests running on a Red Hat Enterprise Linux 5 hypervisor.