What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
You are using the false term "OS Kernel" instead of the correct term "hypervisor or hypervisor kernel" making it seem like you didn't just say "not quite" while agreeing with what I said. DomU, or Xen, or the kernel (all the same thing) is what schedules the CPU. Thereby making it the very definition of bare metal. That it schedules its own tasks or the tasks of its helpers is exactly what bare metal means here.
So now we have different kernels. OK, a kernel does not an OS make. We need a bit of software talking to the kernel to actually be an OS. Now lets see, the DomU is booted up, it loaded a bunch of interfaces that allow software access to the drivers enabling the execution of VMs. So far it's the kernel only. So what is it that started the Dom0, of not a utility software bundled with the DomU?
Don't keep calling the hypervisor an "OS kernel", there is no purpose for that confusion. It's obviously not correct. And there is no reason for it except to avoid stating the obvious... that the hypervisor is what is running on the bare metal. Once you call it what it is, it's clear where it runs.
An OS is a kernel and software utilizing that kernel to whatever purpose. Boot up ESXi, I'm sure you'll find more than just a kernel and a bunch of ABIs there.
There is a reason why semantics is the most important part of communications. Get the semantics accurate, and most confusion tends to go away. It seems that it is only because of non-standard semantics that it seems confusing.
Exactly why I keep saying the common baremetal pitch is wrong
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Relied on, but didn't run on. That's the common myth that started in that era. It was absolutely a type 1 at the time, and we were having these exact discussions at the time about how everyone thought it ran on top of Windows, but didn't. It's been improved since then, but being its own kernel and not running on the Windows one hasn't changed in that time.
It still "relies on" the Windows kernel today, but only in the Dom0. The Windows kernel runs on top of the Hyper-V kernel, and always has.
Can you prove that?
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Type 2 is definitely not as good as type 1 architecturally. More to fail, more layers. Type 2 has its place, but not for production workloads. It's really used for testing. Even there, it is losing popularity quickly.
That's funny, having double the schedulers and double the drivers is better architecturally because..? And since we are there, how can you efine KVM to be type 1 if it uses an OS instead of implementing it's own set of schedulers and interfaces?
It's not better, hence the point. Type 2 needs double the schedulers. Type 2 has the extra layers.
KVM is type 1 because the hypervisor runs on bare metal. The definition is universal, it's not different for each thing. KVM is part of the Linux kernel which, as we established already, is not exclusively an OS kernel, and so KVM does not run on an OS.
KVM is weird to discuss and very confusing because it is polymorphic. You can run KVM without an OS, or with an OS, but in both cases KVM is on the bare metal. KVM is unique in that no other hypervisor kernel is currently capable of being used as an OS kernel. Of course, any kernel could be in you added an OS to it, but no one does, that's silly. And that's why KVM is often seen as bloated, because it has those options and most people use them, at least to some degree. But at the end of the day, KVM is on the bare metal, end of story. And don't say it isn't, because it is. It's on bare metal in the way that the entire industry accepts the term. And it is that use of the term that defines the hypervisor type.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Relied on, but didn't run on. That's the common myth that started in that era. It was absolutely a type 1 at the time, and we were having these exact discussions at the time about how everyone thought it ran on top of Windows, but didn't. It's been improved since then, but being its own kernel and not running on the Windows one hasn't changed in that time.
It still "relies on" the Windows kernel today, but only in the Dom0. The Windows kernel runs on top of the Hyper-V kernel, and always has.
Can you prove that?
Can I prove it beyond the history and documentation and common sense? No. Can you prove that all industry knowledge, records, and vendor information from the time were falsified? I'm not saying it can't happen, I'm saying that it is rather absurd and the position of having the burden of proof lies with the person making the outrageous claim.
This is a discussion that was had back at the time and you are repeating the well dispelled myth that people circulated then. To "prove" what is happening, we need code access. But short of that, it's pretty obvious how it works if you work with it and matches all documentation. If you can prove that you are correct, you've got a lawsuit for MS that you could pull out, or at very least expose closed source software for some pretty serious deception.
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
It's not better, hence the point. Type 2 needs double the schedulers. Type 2 has the extra layers.
No. Xen has schedulers in DomU and then Dom0 enforces additional schedulers of it's own. The same goes in Hyper-V. In KVM there is only one set of schedulers - the ones already existing and perfected over the years in the Linux kernel.
KVM is type 1 because the hypervisor runs on bare metal. The definition is universal, it's not different for each thing. KVM is part of the Linux kernel which, as we established already, is not exclusively an OS kernel, and so KVM does not run on an OS.
KVM is weird to discuss and very confusing because it is polymorphic. You can run KVM without an OS, or with an OS, but in both cases KVM is on the bare metal. KVM is unique in that no other hypervisor kernel is currently capable of being used as an OS kernel. Of course, any kernel could be in you added an OS to it, but no one does, that's silly. And that's why KVM is often seen as bloated, because it has those options and most people use them, at least to some degree. But at the end of the day, KVM is on the bare metal, end of story. And don't say it isn't, because it is. It's on bare metal in the way that the entire industry accepts the term. And it is that use of the term that defines the hypervisor type.
KVM is a kernel module, which requires a kernel and some userspace software (wait, isn't that an OS?) to actually run a VM. You always need those. That's the entire point.
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Can I prove it beyond the history and documentation and common sense? No. Can you prove that all industry knowledge, records, and vendor information from the time were falsified? I'm not saying it can't happen, I'm saying that it is rather absurd and the position of having the burden of proof lies with the person making the outrageous claim.
Can you display those records you mentioned? That documentation?
-
You'll have to excuse me for being skeptical, as claiming that Hyper-V is actually a type 2 (runs on Windows) is the stock example of misconceptions around Hyper-V since day one. This isn't a new claim, it's just one we've heard and seen disproven so many times. And it always ends up being the same things...
- Confusing it with another product like Virtual Server
- Confusing the deployment method with the resulting system
- Confusion caused by the interface
- Confusion based on the use of "physical" to mean a VM by MS
But after going through this hundreds of times, it's always been the same thing. At some point, it's hard to take a new claim seriously.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
KVM is a kernel module, which requires a kernel and some userspace software (wait, isn't that an OS?) to actually run a VM. You always need those. That's the entire point.
You are circling back and ignoring what I've written. No, that's not enough to be an OS. I stated that explicitly a few times to make sure you'd not make this mistake, specifically about KVM. Kernel + "some stuff" isn't an OS on its own.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Can I prove it beyond the history and documentation and common sense? No. Can you prove that all industry knowledge, records, and vendor information from the time were falsified? I'm not saying it can't happen, I'm saying that it is rather absurd and the position of having the burden of proof lies with the person making the outrageous claim.
Can you display those records you mentioned? That documentation?
LIke I said, you need to provide your documentation that goes against everything in the industry. You made the claim.
And yes, just look at Wikipedia. That's the easiest source. You are talking about decade old stuff, just look up years of my writing. I wrote about this at the time, about people making claims as you have.
The burden of proof is on you here.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
It's not better, hence the point. Type 2 needs double the schedulers. Type 2 has the extra layers.
No. Xen has schedulers in DomU and then Dom0 enforces additional schedulers of it's own. The same goes in Hyper-V. In KVM there is only one set of schedulers - the ones already existing and perfected over the years in the Linux kernel.
Dom0 has some schedulers, but that's not really relevant. If you understand type 1 vs type 2, then your statement here doesn't really make sense. I'm not sure what you are trying to say.
-
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
You'll have to excuse me for being skeptical, as claiming that Hyper-V is actually a type 2 (runs on Windows) is the stock example of misconceptions around Hyper-V since day one. This isn't a new claim, it's just one we've heard and seen disproven so many times. And it always ends up being the same things...
I never said it was type 2, types are generally a dumb way of looking at hypervisors. If you have a hypervisor, it is a type 1 by definition, anything else is an emulator.
But after going through this hundreds of times, it's always been the same thing. At some point, it's hard to take a new claim seriously.
It is hard to take a claim of existing proof and documentation without seeing those. You claim something is true - be ready to prove it.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
You'll have to excuse me for being skeptical, as claiming that Hyper-V is actually a type 2 (runs on Windows) is the stock example of misconceptions around Hyper-V since day one. This isn't a new claim, it's just one we've heard and seen disproven so many times. And it always ends up being the same things...
I never said it was type 2, types are generally a dumb way of looking at hypervisors. If you have a hypervisor, it is a type 1 by definition, anything else is an emulator.
There's no real value here. You are just making new definitions. To the rest of us using the industry definitions, there is a lot of value in knowing how these things work. What there is no value in is making up new terms and making there be no way to communicate ideas so that no one can learn about or understand the workings of things.
To you, correct, the value doesn't make sense. But to the rest of us, there is a lot of value in the understanding.
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
You are circling back and ignoring what I've written. No, that's not enough to be an OS. I stated that explicitly a few times to make sure you'd not make this mistake, specifically about KVM. Kernel + "some stuff" isn't an OS on its own.
Sorry but I don't take your word for that, you say it is a mistake, I say it is not. A missile guided by an RTOS or an older phone that could only do a few things, still had operating systems in them, it's just that the scope of those OS's was narrower than that of Linux or Windows. And it isn't "some stuff" (I'll thank you not to put words in my mouth), it is software that utilizes the interfaces the kernel exposes to a specific purpose.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
It is hard to take a claim of existing proof and documentation without seeing those. You claim something is true - be ready to prove it.
I have, for years. You are claiming that those proofs are not true. You are claiming that not only I am wrong, but Wikipedia, Microsoft, and the industry. Yet don't even have a suggestion of supporting documentation. Based on what do you make these wild claims?
I'm simply agreeing that the sky is blue. You are claiming it red. But have nothing to support that theory.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
You are circling back and ignoring what I've written. No, that's not enough to be an OS. I stated that explicitly a few times to make sure you'd not make this mistake, specifically about KVM. Kernel + "some stuff" isn't an OS on its own.
Sorry but I don't take your word for that, you say it is a mistake, I say it is not. A missile guided by an RTOS or an older phone that could only do a few things, still had operating systems in them, it's just that the scope of those OS's was narrower than that of Linux or Windows. And it isn't "some stuff" (I'll thank you not to put words in my mouth), it is software that utilizes the interfaces the kernel exposes to a specific purpose.
Now you are mixing apps with the OS. An RTOS can still do basically anything. As could a phone OS. You are mixing the concept of "general purpose" with the amount of power systems had in the past. Very different concepts.
-
Reference from 2008 era...
https://serverfault.com/questions/326844/is-hyper-v-a-real-hypervisor
When Hyper-V runs as a role on Windows Server 2008 and Windows Server 2008 R2 it's still and hypervisor.
It is type 1 (bare metal) in both cases.
The trick here is that when you install Hyper-V as a role on Windows Server 2008 R2 you see the Windows Server like a host OS and it is not. The setup converts the original OS in something like a VM and puts the hypervisor below. This is what is called the root or parent partition of Hyper-V. That's why you experience the same sped in what you see as the "real machine" and the virtual machines.
Would be extremely fishy for someone to have answered and known how Hyper-V would run in the future, but answered in the past, and got it so exactly correct.
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Dom0 has some schedulers, but that's not really relevant. If you understand type 1 vs type 2, then your statement here doesn't really make sense. I'm not sure what you are trying to say.
Why isn't it relevant? You have a kernel booted up, containing a set of drivers and schedulers, then you have a management VM coming up containing it's own kernel, drivers and schedulers, and some of the system calls a VM executes will have to go through the Dom0's schedulers, to reach the DomU drivers and make syscalls and some will go to the DomU. And yes, some will go to Dom0, which will direct them to DomU (there was a diagram published about all that circa 2011 with the specific calls). How is that more efficient that a single set of schedulers?
You can't just say "this is irrelevant" when it goes against the point you're trying to prove
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
I have, for years. You are claiming that those proofs are not true. You are claiming that not only I am wrong, but Wikipedia, Microsoft, and the industry. Yet don't even have a suggestion of supporting documentation. Based on what do you make these wild claims?
I'm simply agreeing that the sky is blue. You are claiming it red. But have nothing to support that theory.
Oh no, you don't get to turn this one around. You claim hyper-v had the same architecture as xen since its first versions - you prove that.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Dom0 has some schedulers, but that's not really relevant. If you understand type 1 vs type 2, then your statement here doesn't really make sense. I'm not sure what you are trying to say.
Why isn't it relevant? You have a kernel booted up, containing a set of drivers and schedulers, then you have a management VM coming up containing it's own kernel, drivers and schedulers, and some of the system calls a VM executes will have to go through the Dom0's schedulers, to reach the DomU drivers and make syscalls and some will go to the DomU. And yes, some will go to Dom0, which will direct them to DomU (there was a diagram published about all that circa 2011 with the specific calls). How is that more efficient that a single set of schedulers?
You can't just say "this is irrelevant" when it goes against the point you're trying to prove
It's not relevant because it's not under the kernel, so unrelated to what we are discussing. You can layer on as many schedulers on top of things that you want. But we were talking about X and this simply doesn't relate to that discussion.
Your statement of "how is it more efficient than..." doesn't make any sense in this context. It implies something said that wasn't, so there is nothing clear to answer.
