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

[gryning]

I've never seen a good article covering power consumption overheads of the various hypervisors under a comparitive workload.

One common use case is obviously developer laptops, and I do wonder if I switched platform whether I would be better off... then off course the DC bill...

[Gerii]

I've never seen a good article covering power consumption overheads of the various hypervisors under a comparitive workload.

One common use case is obviously developer laptops, and I do wonder if I switched platform whether I would be better off... then off course the DC bill...

I'd really like to know this as well. Just today I used VirtualBox for half an hour or so and my battery percentage dropped considerably faster compared to normal.

[SystemCrasher]

vmWare player is for free

Yet, it does not allows to create VMs if I remember. It only targets use cases where one runs VM prepared by someone else. It not seems to be suitable for benchmarks. And last time I've seen their EULA, it has been prohibiting benchmarking at all. What a scaredy cats they are.

LXD/LXC might be interesting here too.

...but its fundamentally wrong to compare containers vs full blown virtualization software. Containers usually have lightweight overhead on I/O, being very close to bare metal in all regards, randing from I/O up to CPU and memory accesses. Because container is basically thin layer on top of OS kernel which pretends OS is split to several worlds. It only puts very modest overhead and that what makes it interesting.

But containers can't use different kernels for different guests, so it only works as long as guest's user mode software is more or less okay with currently running kernel on the host. You can't have different kernel settings for different guests either. Its some limitation of containers. Also, VMs are somewhat better protected againt malicious attitude of the guest, because in case of containers, attacker basically could do syscalls to host kernel and should there happen some bug, attacker would pwn whole host. VMs are doing it more complicated. Though not entirely impossible, as seen in some fancy recent bugs of Xen and KVM.

[zyrx]

Michael
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).

This. I'm currently running a Xen host and chose the Xen route over KVM as all the anecdotal reports I could find at the time (several years ago) suggested that Xen scaled better. I'd be interested in knowing if this is still the case or if KVM has made some strides here that would make it worth re-investigating. An additional interesting metric as far as Xen is concerned would be to include numbers for a fully virtualized guest as well as its PV counterpart.

[reCAPTCHA]

I don't have a VMware Workstation license, which is why it wasn't tested.

Much more interesting IMO is VMware vSphere Hypervisor, which is free for unlimited number of CPU up to 32 GB of RAM per CPU, though I am not sure about the benchmarking issue. IIRC they don't allow result publishing, so no point than in running it except for personal use...
Another thing is, this is totally different beast compared to Workstation. No graphical interface, and to manage it, one needs Windows vSphere Client application.

[budric]

Much more interesting IMO is VMware vSphere Hypervisor, which is free for unlimited number of CPU up to 32 GB of RAM per CPU, though I am not sure about the benchmarking issue. IIRC they don't allow result publishing, so no point than in running it except for personal use...
Another thing is, this is totally different beast compared to Workstation. No graphical interface, and to manage it, one needs Windows vSphere Client application.

Publish the results under hypervisor: arevmway esxiway

[sarfarazahmad]

well if esxi is here then shouldn't hyper-V be tested too. on a personal experience note, I am witnessing organisations moving away from Hyper-V. KVM clubbed with easy provisioning and configuration mgmt with tools like Foreman are the real cost-saving deals. Yay ! for linux adoption

[SystemCrasher]

well if esxi is here then shouldn't hyper-V be tested too.

Whatever, but I do not come to phoronix to read microsoft marketing bullshit. Since it does not runs on Linux (or at least something *nix-like) and proprietary/vendor locked, it would be better to be benchmarked somewhere else I guess. After all, it runs on windows only. And windows as server is a nightmare.

[drSeehas]

... An additional interesting metric as far as Xen is concerned would be to include numbers for a fully virtualized guest as well as its PV counterpart.

I would add the "PVH" mode to this list (if available in Ubuntu).

[sarfarazahmad]

Whatever, but I do not come to phoronix to read microsoft marketing bullshit. Since it does not runs on Linux (or at least something *nix-like) and proprietary/vendor locked, it would be better to be benchmarked somewhere else I guess. After all, it runs on windows only. And windows as server is a nightmare.

well you got two things right there. Hyper-V and Windows both suck on server platforms. I havent tried out Linux on Hyper-V lately but i did hear about microsoft was contributing code to linux kernel for better supporting GNU/linux on hyper-v/azure.