2 days benchmark KVM,VMware,Xen,Hyper-V
- Thread starter vkeven
- Start date
C
Charlie
Guest
I echo Yatesco's remarks. As much as I prefer to use Proxmox, our initial base test requires we wait for next version of Proxmox and proceed with current Hyper-V plans due to high CPU usage of other hypervisors. The Disk I/O is a key factor for us thus the concentration on disk I/O only for now. We are interested in QCOW2 only for its snapshots and flexibility (Proxmox / KVM), and it was tested againsts Hyper-V's dynamic disk, and the default disk for XenServer 5.6 SP2.
We are using exact same host with same reformatted drive for particular Hypervisor. Guest is plain Windows 2008 R2 SP1 Enterprise. Performed clean install for all setups. See attached - http://tinypic.com/r/2zdr9zo/7 - Hyper-V R2 SP1 and XenServer 5.6 SP2 #47101p and Proxmox 1.8 - these are the latest stable releases as of 07-25-2011.
Comment: VMWare is out since its VSphere Hypervisor has been limited to 8GB for their recent VRAM licensing - will not even consider that. The purpose of this test is to give us some basic comparisons on disk I/O.
We are using exact same host with same reformatted drive for particular Hypervisor. Guest is plain Windows 2008 R2 SP1 Enterprise. Performed clean install for all setups. See attached - http://tinypic.com/r/2zdr9zo/7 - Hyper-V R2 SP1 and XenServer 5.6 SP2 #47101p and Proxmox 1.8 - these are the latest stable releases as of 07-25-2011.
Comment: VMWare is out since its VSphere Hypervisor has been limited to 8GB for their recent VRAM licensing - will not even consider that. The purpose of this test is to give us some basic comparisons on disk I/O.
Last edited by a moderator:
C
Charlie
Guest
Just wanted to update this with additional Proxmox data on RedHat VirtIO driver.
http://tinypic.com/r/i267fd/7
And comparison between Proxmox disk type / driver - both are QCOW2, but first is QEMU IDE and second is VirtIO SCSI
http://tinypic.com/r/2pyxfn5/7 <-- running simultaneously on same system
http://tinypic.com/r/15z0286/7 <-- running individually
The READ appears much faster using RedHat VirtIO than QEMU standard IDE in KVM; however write suffers 2x-4x or more as seen using CrystalDiskMark.
Latest VirtIO drivers from here http://www.linux-kvm.org/page/WindowsGuestDrivers/Download_Drivers were used.
I'd like to see the CPU use go down during intensive I/O and write speed increase using VirtIO in order to compete with Hyper-V.
http://tinypic.com/r/i267fd/7
And comparison between Proxmox disk type / driver - both are QCOW2, but first is QEMU IDE and second is VirtIO SCSI
http://tinypic.com/r/2pyxfn5/7 <-- running simultaneously on same system
http://tinypic.com/r/15z0286/7 <-- running individually
The READ appears much faster using RedHat VirtIO than QEMU standard IDE in KVM; however write suffers 2x-4x or more as seen using CrystalDiskMark.
Latest VirtIO drivers from here http://www.linux-kvm.org/page/WindowsGuestDrivers/Download_Drivers were used.
I'd like to see the CPU use go down during intensive I/O and write speed increase using VirtIO in order to compete with Hyper-V.
Last edited by a moderator:
I have been using the Fedora Project Drivers for a long time on many 2008 R2 VMs.
They are signed and work fine but they are not WHQL as the above URL states.
I only use the virtio disk drivers because I keep seeing people having random issues with lan drivers.
Even a few Debian Squeeze guests I setup had issues when using virtio lan.
The Lan is fast enough for my needs using e1000 but I agree it would be great to see better performance.
C
Charlie
Guest
Same here. I only use the disk drivers. I also use the e1000 with the updated version from Intel.