Ubuntu 15.10: KVM vs. Xen vs. VirtualBox Virtualization Performance

[Xaero_Vincent]

This is what's mentioned in the universal EULA:

2.4 Benchmarking. You may use the Software to conduct internal performance testing and benchmarking studies. You may only publish or otherwise distribute the results of such studies to third parties as follows: (a) if with respect to VMware?s Workstation or Fusion products, only if You provide a copy of Your study to [email protected] prior to distribution; (b) if with respect to any other Software, only if VMware has reviewed and approved of the methodology, assumptions and other parameters of the study (please contact VMware at [email protected] to request such review and approval) prior to such publication and distribution.

[ruthan]

Where is vmware?

[killyou]

I recently migrated from VirtualBox to KVM and I noticed that my host CPU usage dropped significantly and that's crucial for me since I run 10 virtual routers all the time for the studying purposes. What annoyed me in VirtualBox is that sometimes after starting a VM it was consuming 25% of the host CPU no matter what while after a guest restart it could drop to as low as 1%.

[chepati]

What I almost never see in these type of comparisons is network performance. I've been running a firewall inside first VirtualBox, then KVM. My gut feeling is that KVM does better with the virtio net driver than VirtualBox, but I can't scientifically substantiate it. I have no idea how Xen measures in that regard.

And even though it matters little to me, I'm sure plenty other readers would be interested in 2D and 3D numbers.

Michael, please include such tests if possible in this benchmarking round, or at least keep it mind next time. Thanks.

[nils_]

I sometimes wore a sysadmin hat at my last job, and for KVM disk I/O, using the "VirtIO" emulated disk format made performance awesome versus using the emulated IDE disks - very close to native. I don't know if VirtIO is the default for KVM on 15.10 or not.

Yeah I'm getting pretty good results using virtio-scsi on a raw lvm volume, and at least decent performance on ceph (latency sucks though).

[darrent]

Michael
I have been "lurking" here for some time and have appreciated your articles and original content. (yes I'm considering taking up a subscription).

The article is great in that it gives a comparison of a single VM against the same machine running "bare metal", but frankly that is interesting but not "useful"
What I would find more useful is how well each of the VM platforms "scales out" as more VM's are added - as that is more like how I'm using the virtualisation platforms.

e.g. what is the "average performance" of 1x, 2x, 4x, 8x, up to ...{N)x VM's ( e.g. a 4GB RAM, dual CPU VM, sharing and contending for access to the host resources).

I believe that I have some groups of VM's running where the "total aggregate performance of the group" is actually better that would have been achieved on a single OS instance on the same hardware. it would be great to have some "numbers" to measure that against....

Keep up the good work

Regards
Darren

[inhuman4]

LXD/LXC might be interesting here too.

Agreed. I'm currently using KVM for windows VMs and LXC for linux VMs. I'd be very interested in seeing how the two technologies affect disk access. LXC should be bare metal speed but it would be nice to know for sure.

[Passso]

This is what's mentioned in the universal EULA:

2.4 Benchmarking. You may use the Software to conduct internal performance testing and benchmarking studies. You may only publish or otherwise distribute the results of such studies to third parties as follows: (a) if with respect to VMware’s Workstation or Fusion products, only if You provide a copy of Your study to [email protected] prior to distribution; (b) if with respect to any other Software, only if VMware has reviewed and approved of the methodology, assumptions and other parameters of the study (please contact VMware at [email protected] to request such review and approval) prior to such publication and distribution.

*** You may benchmark VmWare only if you send us the results first so that we check it beats other virtualizers. ***

Now I understand why I cannot find any benchmark of VmWare on the net...

[Slartifartblast]

*** You may benchmark VmWare only if you send us the results first so that we check it beats other virtualizers. ***

Now I understand why I cannot find any benchmark of VmWare on the net...

Ha, laughable. See you in court VMware as that EULA clause doesn't have a legal leg to stand on.

[bug77]

I sometimes wore a sysadmin hat at my last job, and for KVM disk I/O, using the "VirtIO" emulated disk format made performance awesome versus using the emulated IDE disks - very close to native. I don't know if VirtIO is the default for KVM on 15.10 or not.



That used to be in their EULA, I don't know if it still is because I haven't used VMware in years.

It's been standing very well for years.
The downside (for VMware) is they're automatically excluded from any public comparison. And that translates into little to no word of mouth.