Ask HN: What hypervisor are you using?

Having used, extensively and in anger, Hyper-V 2008/R2/2012, VMWare ESXi 4 thru 5.5 and (Citrix) XenServer, for Windows and Linux VM's, my personal favourite is XenServer.

Hyper-V works, but has terrible admin tooling, especially if not on a domain

VMWare is a fine hypervisor and if you are happy with what it provides then ESXi free is hard to beat

But if you want a cluster on the cheap, with proper clustering support, XenServer is hard to beat. I use this extensively especially in SMB and also local dev environments - a couple of Xen Servers backed by fastish NFS SAN's. Easy to maintain and administer. Good tooling both from a GUI and CLI point of view, without actually having to dish out any money.

That said, if someone else was paying, ESX would be just as good if you needed vMotion/vStorageMotion etc.

Agreed on the horrible Hyper-V tooling.

I still use it begrudgingly though, because my dev machine is also a gaming machine, and Hyper-V GPU passthrough to the primary OS (Windows) is nigh seamless and doesn't require any configuration. Also, my dev machine also happens to be a laptop, and Hyper-V supports many advanced power management features (like connected standby), that are not well-supported on any other hypervisor as far as I'm aware of (many of them have trouble with even going into/resuming from sleep/hibernation in the first place when I last tried them).

The difference between Hyper-V tooling on a domain vs off a domain is really shocking. The tool sets you get on a domain can be useful...off a domain...I'll pass entirely.

I've run barebones hypervisors in production for years; KVM is head and shoulders above everything else I've tried. libvirt and virsh provide a reasonably nice interface if you don't need web-based tooling, and of course config management makes everything a lot easier to maintain. I haven't used salt, but if it has mature tooling for what you need, KVM would be a no-brainer for me.

oVirt is the web interface for libvirt stuff. There is also virt-manager for a desktop GUI for libvirt stuff.

ganeti is an alternative to libvirt.

At https://mnx.io we are using http://smartos.org from the great folks at http://Joyent.com

You can look at https://project-fifo.net for managing smartos.

I'd be happy to discuss in detail if it looks interesting!

I use libguestfs and virsh to inject files in a VM (without network).

http://libvirt.org/virshcmdref.html http://libguestfs.org/

We run KVM on all of our current machines (about 20-25 VMs at any given time), managed by Cloudmin (eating our own dog food). Before that, it was Xen, which also worked well. The decision is mostly one of least resistance; KVM has the best out of the box support on CentOS 7, so it's what we're using. Back when Xen met that description, we used Xen.

I have used KVM. I have a standard Ubuntu workstation, but sometimes want to do something like have a Debian sid system building the trunk version of Gnome via jhbuild or something. If it falls apart, my base system is not affected.

I have VPS's at Linode and Ramnode. Both these companies use KVM as well. Even if you're not deploying cloud, you might look at what the big VPS providers are using, as they're dependent on a sturdy hypervisor, so big VPS providers using a particular one is a good indicator.

I have worked at places that used VMware's hypervisors, and they worked fine. You'll be shelling out more money for the support and such though, although they do have some free options on the lower end. I am not up to date on VMware enough to know if there's any catches on the free VMware stuff now in 2016.

SmartOS is a good choice. They provide KVM and joyent opensource their SmartDataCenter https://github.com/joyent/sdc

so you can provision VM with API.

They are developing for LXC which will give you better performance in the future.

...Wait. Did you just say they're developing LXC? That is hilarious.

They're SmartOS! They're on the Solaris codebase! They've had containers since 2005! Literally before it existed, it had containers!

linc01n's comment was a little unclear. SmartOS isn't doing anything with the LXC protect. What SmartOS has done is revived the branded zones code from Sun for emulating Linux system calls. This means that you can run Docker containers on SmartOS as zones, as well as use dtrace, mdb and ptools on your Linux binaries. You can read up some more here:

https://wiki.smartos.org/display/DOC/LX+Branded+Zones

I saw Bryan Cantrill's talk on that! It did sound neat.

I use Virtual Box, but it is based on QEMU.

I used QEMU sometimes because it has support for MIPS and SPARC emulation so that it can run old operating systems for those platforms on it.

I had a task to modify QEMU to get networking working on a Solaris 7 SPARC virtual machine, but there is a bug in the kernel that causes a crash with the NIC that is emulated in QEMU for SPARC and there used to be a Sun Y2K CD-ROM that fixed it and nobody seems to have it anymore. Anyway I couldn't modify QEMU in time to make it work and lost that gig. To be honest there are Russian Hackers trying to get Solaris 7 and older stuff working in QEMU and can't seem to get anywhere. But that was a few years ago and maybe someone made progress yet?

QEMU runs IBM OS/2 quite well.

KVM. SolusVM on top for clients / people who don't want to screw with the command line.

We've been using Proxmox VE ( https://pve.proxmox.com/wiki/Main_Page ) in production for a while.

It's open source and free, and is a tweaked Debian with KVM/QEMU virtualization and also LXC container support. It puts a nice web interface on top of it all, but the underlying OS and CLI is fully available too.

The hosts can be clustered and VMs can be migrated. It doesn't require shared storage but can use it if you have it. (You need shared storage for live migrations).

We pay for support since the cost is reasonable, but you don't have to.

Is there a reason you went with that option? I've never heard of Proxmox VE so I'm curious to hear why you went that route vs just plain KVM.

Depending on your needs, LXC might fit the bill. Otherwise, KVM.

Somewhat related: anyone know if the "API" for ESXi/vSphere works on the free version?

I've considered putting the free hypervisor on a server here at home instead of using VirtualBox on my laptop for my test VMs but I'd want to be able to create/provision/etc. VMs using my own tools (and not have to fire up a Windows machine to run the vSphere client).

As of a few months ago, no. They only open up the API when you license it. They may change it in the future as they no longer have a strangle hold on the hypervisor market, but traditionally, they withhold that API functionality in the free version.

Xen, but only because its the primary target of pretty much every current unikernel build/deploy chain.

As that becomes less and less the case we may migrate or expand what we adopt.

Have you checked with hyper.sh?

KVM is King(Linux/Openstack), Xen is dead, Docker is unstable.

Docker isn't a hypervisor...

Maybe that's why he thinks Docker is unstable.

Looks like KVM has poor support for PCI passthrough for GPU (e.g. OpenGL and CUDA)

I disagree. KVM has the best and most flexible passthough support that I've used.

Successfully running Windows 10 on top of Arch, with my GTX 970 passed though. Works without a hitch.

>Xen is dead

Other than that whole AWS thing...

And IBM/SoftLayer...

Which doesn't really count. Its fine AWS uses Xen, they have the budget and the team to maintain their fork, warts and all.