TR Forums

I'm curious to poll some thoughts on the main three options, or if there's anything new out there? Are there any particular strengths or weaknesses in any of them that decided which one you use? I left out Parallels as that's Mac-only now I believe.

I've had extensive use with VB a few years ago before Oracle bought them, and VMware since. I've been tempted to try HyperV since it's free on Win 8 Pro and is the only one to actually support VT-d to date, but given the recent mouse issues (and that my 4770K lacks VTd anyway) I've held off upgrading the OS for the time being. But I'm curious to hear what people think of the new Hyper-V in particular given I've yet to use it.

If I didn't have to worry about paying for it, VMware. But for my personal use, VirtualBox is sufficient.

phpvbox + cheap AMD bulldozer + vagrant = cheap playground for crazy systems integration ideas. Hyper-v, VMware are the real platform for building out a serious VM server, but who want's to do that? At work we develop things as disposable VMs and then put them in a cheap cloud provider anyway. Virtualbox has lots of cool things that work with it because it's free as in beer.

Guess it really depends on what you want to do with your VM. If you are planning on utilizing for anything that requires the best graphics performance available in a VM, or proper USB passthru implementation then VMware is really your only choice.

If you just need something to quickly test run development code, VB should suffice unless again you need proper USB or video acceleration (BTW VT-d IS supported on VB for certain NIC cards).

As far as Hyper-V and VT-d goes, I guess my question would be to you would be what types of devices are you planning to work with VT-d. My experience has been that there are very few consumer products that play well with it. So unless you are planning to have isolated NICs or controller cards I don't really see any advantage going the Hyper-V route.

Deanjo wrote:

Guess it really depends on what you want to do with your VM. If you are planning on utilizing for anything that requires the best graphics performance available in a VM, or proper USB passthru implementation then VMware is really your only choice.

I got into VMs back when SMP F@H folding flourished under linux code. So everyone was using 2-4 linux VMs to maximize the PPD out those shiny new Core 2 Quad chips! Around 2007-ish? But currently I'm playing around with VMs with mixed processor & graphics rendering loads.

The VM's I play around with are "disposable" as GodsMadClown phrased it. I made one XP VM and cloned it repeatedly.... linked clones are especially awesome btw! But every time I break one I can just delete and re-clone another.

Deanjo wrote:

As far as Hyper-V and VT-d goes, I guess my question would be to you would be what types of devices are you planning to work with VT-d. My experience has been that there are very few consumer products that play well with it. So unless you are planning to have isolated NICs or controller cards I don't really see any advantage going the Hyper-V route.

The resulting loads I'm running will 100% max out my 4770K culminating with around 80-90% red kernel overhead. Eventually the VMs begin to bog down before something breaks from the abuse in the VM software My understanding is VT-d will reduce a lot of the IO overhead involved in VM GPU processing as well as system IO in general. It is something I plan to test the moment I have a VT-d capable processor as I'm incredibly curious to see what sort of difference, if any, that it would make. Opinions tend to vary wildly when it comes to VT-d so that's an entire thread of its own, but I'm mostly during this for fun and simple curiosity.

Most Z87 boards appear to be properly equipped for VT-d, at least with GB there's the basic VTx UEFI setting and then a second, VTd specific setting as well in the UEFI.

If we are talking about "processor and graphics workloads", this is 2013 and I am fairly certain that most of these applications should be multithreaded by now that utilizing VMs for "distributed-ness" is actually worse? The 80-90% red kernel overhead suggested that all those VMs that you are running is causing all sorts of context switches either via I/O loads or just the VM processes juggling for CPU attention. This does not sound like you are doing any tangible work (you want non-kernel overhead to be closer to 100% utilization instead of the OS spending most of its time spinning its wheels going in and out of kernel mode).

I have to ask, is this idea of running multiple VMs even a good idea?

I use VMWare Player, which is free for personal use, to run an XP machine on Win 7.
I did use Virtualbox for a while, a few years ago, during a Linux experiment. With that said, I find that VMWare has better hardware acceleration over Virtualbox.

If you have Windows 8 or 8.1 Pro, I'd use Hyper-V over any of the other free solutions. IME it knocks the spots of VirtualBox in terms of speed and reliability. The only thing it lacks is support for 3D acceleration inside a VM. Even Linux support is improving too although I suspect VB and VMware will still be better for that.

Otherwise I'd considering paying for VMware Workstation.

Been using VirtualBox for a few years on both Windows and Linux hosts. I realize that VMware does some things better, but VirtualBox is good enough, free, and I'm familiar with it.

New owners of the company I work for require that everyone's primary desktop run Windows 7 Enterprise. I' m more of a Linux guy (and develop stuff for Linux targets), so I guess I will be making even more use of VirtualBox going forward. I've already got a pretty nice Kubuntu 12.04 VM set up that I am migrating most of my day-to-day development work to.

I use VirtualBox for Linux guests because it's a lot simpler to install the guest utilities than for VMware (which has to be untarred and run manually), and for archaic Linuxes like Debian 2.2 it has somewhat better-supported emulated hardware. For Windows guests I could go either way since I don't run 3D apps, hence VMware's greater speed doesn't do much for me, however it seems like VMware's been somewhat less crashy with Windows guests over the past several versions of both it and VB.

My Windows 98 VM (for certain old games, like SimCity 2000) has to run on VMware because there aren't any '98 guest utils in VirtualBox, likewise my bootleg OS X virtual machine.

I regularly use VirtualBox on Windows, Linux and OS X hosts running Windows, Linux and BSD guests. Never had a problem, always solid, minimal maintenance.

In the past, I've used VMWare Fusion on an OS X host running a Linux and Windows guests, VMWare Player on a Windows host running Linux guests, and ESXi on a dedicated box that ran two BSD guests. In my experience, the VMWare setups always needed more maintenance, had more overhead, less efficient memory usage, and were more invasive to the Host. The ESXi box did not age well and started to need regular reboots to maintain full performance of the guest VMs, though there were no hardware or memory issues in that box.

If I were running many hosts, or needed good remote monitoring, and if cost were not an issue, I'd probably (still) go with VMWare, but for 2 or 3 hosts, each running 2 or 3 guests, VirtualBox has been terrific.

VirtualBox for applications that don't work on 64-bit Windows, and Hyper-V for everything else. The Server 2012 R2 Hyper-V is teh r0x0rs.

I dont quite have a favorite I have used them all. However, my current dev stuff resides in Hyper V and ESX.

bthylafh wrote:

I use VirtualBox for Linux guests because it's a lot simpler to install the guest utilities than for VMware (which has to be untarred and run manually), and for archaic Linuxes like Debian 2.2 it has somewhat better-supported emulated hardware.

VMware has done a good job of working with the community to provide guest drivers for inclusion in Linux, and they have repos for RHEL/CentOS, Ubuntu, and Suse. Recent versions of RHEL/CentOS/SL already have the open source VMware drivers integrated.

Open Virtual Machine Tools:
http://open-vm-tools.sourceforge.net/index.php

VMware Operating System Specific Packages (OSPs):
http://www.vmware.com/support/packages

Flying Fox wrote:

If we are talking about "processor and graphics workloads", this is 2013 and I am fairly certain that most of these applications should be multithreaded by now that utilizing VMs for "distributed-ness" is actually worse?

Back when Pande Group first introduced its SMP folding program (2007ish), the SMP F@H Windows code was in bad shape and not optimized for Windows. Folders using the Q6600's would see something like a 20% increase in performance if they ran a pair of linux dualcore VMs inside their Windows box and folded inside each of those, assuming they didn't just run linux natively. That should illustrate just how bad the Windows version was at first. These days thankfully they are much more on top of their software development and users are better off running native regardless of OS. I just cited it as what got me into virtual machines in the first place!

Flying Fox wrote:

The 80-90% red kernel overhead suggested that all those VMs that you are running is causing all sorts of context switches either via I/O loads or just the VM processes juggling for CPU attention.

So do you think VTd would help in that instance or not really? Will be awhile before I can test this personally.

Flying Fox wrote:

I have to ask, is this idea of running multiple VMs even a good idea?

Probably not, but again I'm doing it just for fun. And this thread is just for curiosity's sake! I have already used VB and Vmware both, and plan to try Hyper-V just so I can get a feel for it and be familiar with all three. But I'm not a programmer or software dev, and I only know enough about Linux to be dangerous (I've managed to break a few linux distros mucking around in them without knowing what I was doing ). So I thought it'd be neat to get some more enlightened & experienced opinions to go with my own for all the VM programs. I do appreciate all the responses!

Kougar wrote:

but again I'm doing it just for fun.

I submit that soon, the idea of virus scanners/checkers will go away (having been defeated) in favor of disposable VMs that replace today's concept of "browse in incognito mode". With sandboxing, it's starting to be this way already, but soon opening a browser, or any other application, will be the same thing as starting up a VM.

Efforts like Qubes-OS (qubes-os.org) have started to dance around this idea, but still don't quite go far enough.

MarkG509 wrote:

I submit that soon, the idea of virus scanners/checkers will go away (having been defeated) in favor of disposable VMs that replace today's concept of "browse in incognito mode". With sandboxing, it's starting to be this way already, but soon opening a browser, or any other application, will be the same thing as starting up a VM.

Efforts like Qubes-OS (qubes-os.org) have started to dance around this idea, but still don't quite go far enough.

There will always be the need to protect the host systems, especially as those can infect a guest system. But otherwise yeah... why not? When it just takes a couple clicks to clone an entire system that's already been configured, migrating to disposable VMs only makes sense. Modern desktops are capable of running a dozen loaded VM systems at once inside a host OS. I've also read about security researchers and others using VMs to test software packages for malware, and others using them for private browsing. It certainly takes any uncertainty out of the equation.

I do agree that the PC of the future will be heavily virtualized, if only for the fact that users will need legacy applications after Microsoft goes full mobile forcing everyone onto *nix. From what I've seen, VMware offers the best performance and is very reasonably priced for business-class software.

Kougar wrote:

But otherwise yeah... why not?

In a word, "benchies". I think we need a benchmark for benchmarks about how well they reflect and affect real-world performance. E.g., most browser benchmarks focus on things like time to open a web page, but really >90% of the time people see/feel in real life is network latency. Personally, I can "feel" 3ms, "see" 10ms, but don't care much about anything less than 100ms for desktop stuff (and right, I don't game (much).) That's plenty of time to slide virtualization underneath most apps.

I have a few weeks of use-it-or-lose-it vacation coming up. I'm thinking of taking the VirtualBox source plus the Firefox source, and doing something like sticking them together into a "FireBox". Might even add in pieces of Truecrypt source.

MarkG509 wrote:

In a word, "benchies". I think we need a benchmark for benchmarks about how well they reflect and affect real-world performance. E.g., most browser benchmarks focus on things like time to open a web page, but really >90% of the time people see/feel in real life is network latency. Personally, I can "feel" 3ms, "see" 10ms, but don't care much about anything less than 100ms for desktop stuff (and right, I don't game (much).)

It's interesting ya should give that example. With broadband the only time I pay attention to page loads is when some plugin is making the browser sap a full CPU core. Or when I'm running the GPU Folding client. Ever since most browsers adopted GPU acceleration it now will affect page load times noticeably... and any page with flash content becomes nearly unbrowsable. Yet if I watch a movie instead I don't notice an issue as long as I don't interact outside VLC. I think there's still room to improve on "hyperthreading" the GPU for handling multiple concurrent programs, to badly misuse the term.

More toward you point, that's an interesting idea to run with. I'm sure you'd gain some interest from users wanting ultra-private or an additional level of security with their browser! If you do go for it I might suggest testing it with the host GPU loaded with something else and making sure it doesn't affect the Guest firefox load time? Although latency speaking nothing really loads a GPU as severely as F@H will so it probably won't matter much.

Kougar wrote:

I think there's still room to improve on "hyperthreading" the GPU for handling multiple concurrent programs, to badly misuse the term..

I get your drift. The problem is that the "state" is way over there, on the other side of the PCIe bus, in it's own local memory that makes sharing GPUs is a problem, and carving up GPUs requires cooperation that the various clients don't know they need to do. I really like idea of the shared embedded DRAM, that also works like an L4 cache, in the Intel "Iris Pro" graphics adapters.

Popping a digression or two off the stack, I think we're circling around a good point, which is what is the right amount of state to virtualize, and what Virtual Machine Managers can do, or try to do, to help with this. When I first create a VM, it really is throw-away. But after a few days, when I've set it up with all my 'stuff' it starts to become as important to me as the physical machines on which it's running. That tells me I'm doing something wrong by virtualizing the whole thing. Add Windows into the picture, and on top of the state issue, there's now s/w license(s) locked to the VM.

Yeah, the product activation conundrum does place some limits on what you can do with Windows VMs unless you've got a VLK. At least you still have the 30 day window if using a standard retail/OEM key.

With Linux I routinely move installs back and forth between VMs and real hardware. It's pretty cool to be able to set up a system in a VM, then dump the image to a physical hard drive and boot it natively.

real men use KVM, anything else is amateur hour.

shank15217 wrote:

real men use KVM, anything else is amateur hour.

Yeah, yeah... and you're only a "real" Linux user if you do LFS (or at least Gentoo), right?

just brew it! wrote:

With Linux I routinely move installs back and forth between VMs and real hardware. It's pretty cool to be able to set up a system in a VM, then dump the image to a physical hard drive and boot it natively.

Never quite tried that, but I carry around a USB 3 drive with copies of the virtual machines' disk image files. Use it to move them to different physical machines, or restore after a some failure. Three times in the last year, that saved my bacon. There's just nothing quite a sweet as copying 1 file (albeit ~100GB sometimes) to get back to a known/working/clean state.

MarkG509 wrote:

Popping a digression or two off the stack, I think we're circling around a good point, which is what is the right amount of state to virtualize, and what Virtual Machine Managers can do, or try to do, to help with this. When I first create a VM, it really is throw-away. But after a few days, when I've set it up with all my 'stuff' it starts to become as important to me as the physical machines on which it's running. That tells me I'm doing something wrong by virtualizing the whole thing.

Well given the amount of time it takes to configure and optimize the Guest OS and programs used in it, it only makes sense nobody would want to see all that time thrown out the window! Of course, once you virtualize a host OS ya can just copy it for backup purposes and re-image if the system ever has a problem!

just brew it! wrote:

Yeah, the product activation conundrum does place some limits on what you can do with Windows VMs unless you've got a VLK. At least you still have the 30 day window if using a standard retail/OEM key.

Yeah, the Windows licensing issue is annoying... if for example someone bought Windows 8 then it would be nice if they could run personal, non-commercial VM's of it within itself. I'd like to try out some Win 8.1 VMs just to feel out how they compare against my XP VMs, but setting one up just to toss it has been enough to deter me so far. Was why I was sad to see TechNet go away as that was perfect for just testing and playing around with windows VMs.

Just out of curiosity what are the rules pertaining to Windows being installed in a VM?

AFAIK it counts just like any other Windows installation. A VM is considered a separate "system" for licensing purposes.

What about moving VMs, will Windows detect the changed motherboard and require re-activation?

NovusBogus wrote:

What about moving VMs, will Windows detect the changed motherboard and require re-activation?

Yea thats what I was wondering about. I can see scenarios where you might want to move a VM between computers or even clone one as a form of backup or to refresh your environment back to a known state.