| |||||||||||||
|
| |||||||||||||
Red Hat acquires QumranetRed Hat acquires QumranetPosted Sep 4, 2008 19:20 UTC (Thu) by hisdad (subscriber, #5375)Parent article: Red Hat acquires Qumranet
Its not the coupling between the kernel and the kvm that's the issue.
Its the coupling between the guest os and the hypervisor.
How much do you patch your guest to get good performance??
(Log in to post comments)
Red Hat acquires Qumranet Posted Sep 5, 2008 2:40 UTC (Fri) by drag (subscriber, #31333) [Link]
Well the code you modify in the guest will be in the form of paravirtualized drivers.
So for Linux clients they are slowly integrating PV drivers as KVM matures. So there are PV drivers for block devices and network drivers and memory balloon drivers (for reclaiming memory from VMs.
http://kvm.qumranet.com/kvmwiki/Virtio
For example the fastest non-PV device for KVM is the Intel 1Gb/s ethernet adapter. When I moved to a PV driver increased the network performance of the virtual network by 300% and lowered the cpu usage in both the guest and host.
Then there is only a network driver for Windows for the same thing.
Not only they are faster Posted Sep 10, 2008 17:05 UTC (Wed) by gwolf (subscriber, #14632) [Link]
As virtio drivers don't have to interact with real-life hardware, only forward the requests to the hardware interfaces in the host OS, virtio drivers are also much simpler - they will probably be more secure.
|
|
||||||||||||