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:
Type 1 hypervisor means that the hypervisor itself runs without being "managed" by another layer of code, it is the "controlling" kernel that controls access to the CPU through scheduling.
A Type 2 hypervisor gets scheduled by the OS it is running on top of. A type 2 hypervisor relies on the OS kernel that you describe.
I still don't like the description (and the fact you managed to twist it quite a bit). The official description of the types is that type 1 works directly with the hardware without an assisting OS (which is absurd) and type 2 needs to go through an OS before it gets to the hardware. What you did here is basically remove the CPU extensions drivers and start talking about schedulers. Folks who don't know any better might even think type 2 (the way you describe it) isn't as good as type 1, because "schedulers".
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
The early versions of hyper-v were loaded on top of windows actually. Later in the game, MS got tired of lagging behind so they hired some of the Xen folks to do a complete architectural rewrite.
I think you are thinking of Microsoft Virtual Server, not Hyper-V. Virtual Server was a type 2 hypervisor from MS that they used prior to Hyper-V. Hyper-V was bare metal (aka native, aka Type 1) from the start (2008.) While MS can apply the same name to two unrelated products (Windows 95 and Windows 2000), they rarely do and I am pretty sure that they did not with Hyper-V. Hyper-V was type 1 on its first day on the market.
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Right, and this is what is wrong. If they did what you are describing, that's not bare metal, that's hardware virtualization (aka, IBM, Oracle, HPE implementation.)
And it isn't available on commodity hardware.
There IS true bare metal. It's how we all do it. You could argue that Xen and Hyper-V aren't bare metal, but that requires redefining bare metal hypervisors to do so. But KVM and ESXi have no possibility but being bare metal, there is nothing beneath them at all, and no helpers anywhere.
Not quite. There is direct access to the CPU extensions, scheduled by the OS kernel (even if you prefer to call that OS vmkernel or DomU). And there is a ton of supporting software emulated stuff attached, because a VM is more than just a CPU.
-
@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:
Type 1 hypervisor means that the hypervisor itself runs without being "managed" by another layer of code, it is the "controlling" kernel that controls access to the CPU through scheduling.
A Type 2 hypervisor gets scheduled by the OS it is running on top of. A type 2 hypervisor relies on the OS kernel that you describe.
I still don't like the description (and the fact you managed to twist it quite a bit). The official description of the types is that type 1 works directly with the hardware without an assisting OS (which is absurd) and type 2 needs to go through an OS before it gets to the hardware. What you did here is basically remove the CPU extensions drivers and start talking about schedulers. Folks who don't know any better might even think type 2 (the way you describe it) isn't as good as type 1, because "schedulers".
I've twisted nothing. And you aren't repeating that definition accurately. It's close, but modified just enough to make it not make sense. Type 1 runs directly ON the hardware, and there is nothing about if an assisting OS exists (that would be weird and makes no sense, hence your confusion.) If you use the standard definitions, it's all clear and sensible. It's your modification of them that makes it all seem crazy.
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.
I've not removed the CPU extensions, because they have never been relevant. They aren't part of any definition, and aren't needed for the conversation. You bring them up and by adding them in as part of your definitions make the simple, straightforward, and sensible definitions that everyone accepts and has accepted for decades seem crazy. If you don't inject a need for them into the definitions, suddenly the standard definitions are totally logical.
CPU extensions are great, but at the end of the day, they are "helpers" and remain optional. A good option, but an option nonetheless.
-
@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:
Right, and this is what is wrong. If they did what you are describing, that's not bare metal, that's hardware virtualization (aka, IBM, Oracle, HPE implementation.)
And it isn't available on commodity hardware.
Sure. Because we define commodity hardware as not having that feature. Add that feature, to anything, and we stop calling it commodity.
It isn't like virtualization is unique to commodity hardware. In the past, it was the one type that didn't have virtualization (hence why that market was left with a knowledge cap leading to confusion when it was suddenly introduced to something that the rest of IT had long ago standardized on and accepted.)
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
I dont' agree here at all. People try to make terms fluid for some reason, but they are not. It's not loose English meaning "whatever people who don't know use them to mean", it is technical jargon and well known, documented, and clear.
Baremetal is a very well established term and I've never once heard anyone not know what it meant before. Not across any community or media. It's not non-sense and is an extremely important term.
It is nonsense in the way it is peddled by the major vendors' marketing teams. What they keep pouring into people's ears is what I've been explaining to the more junior IT folks to be wrong - baremetal is not software-free, on-hardware. Hypervisor types do not make sense (because if you ask a vmware salesman you'll hear the baremetal-not-baremetal argument right there). Nothing works on the hardware directly, everything goes through layers. A VM you start has to go through a layer to get CPU time, luckily it's just one layer, but you also need to get scheduled inside and outside the VM, so you'll have overhead (and gang scheduling if you paid for vmware
). In short, there is no magic in virtualization, and saying something is "baremetal" does not make your VMs actually run on the real hardware. -
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
There IS true bare metal. It's how we all do it. You could argue that Xen and Hyper-V aren't bare metal, but that requires redefining bare metal hypervisors to do so. But KVM and ESXi have no possibility but being bare metal, there is nothing beneath them at all, and no helpers anywhere.
Not quite. There is direct access to the CPU extensions, scheduled by the OS kernel (even if you prefer to call that OS vmkernel or DomU). And there is a ton of supporting software emulated stuff attached, because a VM is more than just a CPU.
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.
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.
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.
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
I think you are thinking of Microsoft Virtual Server, not Hyper-V. Virtual Server was a type 2 hypervisor from MS that they used prior to Hyper-V. Hyper-V was bare metal (aka native, aka Type 1) from the start (2008.) While MS can apply the same name to two unrelated products (Windows 95 and Windows 2000), they rarely do and I am pretty sure that they did not with Hyper-V. Hyper-V was type 1 on its first day on the market.
Nope, Hyper-V on 2008 and 2008R2 relied on the windows kernel, in 2012 it got separated into a separate kernel. Xen folks managed the overhaul, don't remember the names right now.
-
@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:
I dont' agree here at all. People try to make terms fluid for some reason, but they are not. It's not loose English meaning "whatever people who don't know use them to mean", it is technical jargon and well known, documented, and clear.
Baremetal is a very well established term and I've never once heard anyone not know what it meant before. Not across any community or media. It's not non-sense and is an extremely important term.
It is nonsense in the way it is peddled by the major vendors' marketing teams. What they keep pouring into people's ears is what I've been explaining to the more junior IT folks to be wrong - baremetal is not software-free, on-hardware.
What vendor, even in marketing, is saying this? I'm not disagreeing with you, but this goes against all experience that I've had. I've never heard anyone hint and there being junior IT folks making this crazy mistake.
I've seen some insanity out there with what people don't understand, so I totally understand that this would be possible. But I see two issues here...
- I'm not confident that this is a widespread issue. Look at ML or SW, even on SW I've never heard of a single person making this particular mistake.
- You are arguing that something isn't bare metal to people who actually know what it is, based on the assumption we are using the term to mean something very different than what it means.
If someone IS having this confusion, I would warn against assuming it is common. Maybe it is and somehow ML, SW, and all IT walks of life I've seen have avoided it in a bubble, that can happen. but I talk to a lot of people and have seen a lot of crazy misconceptions. This feels like something I'd have been extremely likely to have seen if it was happening much at all. Maybe Reddit is worse than SW and doing 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:
I think you are thinking of Microsoft Virtual Server, not Hyper-V. Virtual Server was a type 2 hypervisor from MS that they used prior to Hyper-V. Hyper-V was bare metal (aka native, aka Type 1) from the start (2008.) While MS can apply the same name to two unrelated products (Windows 95 and Windows 2000), they rarely do and I am pretty sure that they did not with Hyper-V. Hyper-V was type 1 on its first day on the market.
Nope, Hyper-V on 2008 and 2008R2 relied on the windows kernel, in 2012 it got separated into a separate kernel. Xen folks managed the overhaul, don't remember the names right now.
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.
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
I've twisted nothing. And you aren't repeating that definition accurately. It's close, but modified just enough to make it not make sense. Type 1 runs directly ON the hardware, and there is nothing about if an assisting OS exists (that would be weird and makes no sense, hence your confusion.) If you use the standard definitions, it's all clear and sensible. It's your modification of them that makes it all seem crazy.
Wait, you're saying an assisting OS doesn't exist? What is the DomU then? It is way more than just a driver for VT/SVM.
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?
I've not removed the CPU extensions, because they have never been relevant. They aren't part of any definition, and aren't needed for the conversation. You bring them up and by adding them in as part of your definitions make the simple, straightforward, and sensible definitions that everyone accepts and has accepted for decades seem crazy. If you don't inject a need for them into the definitions, suddenly the standard definitions are totally logical.
CPU extensions are great, but at the end of the day, they are "helpers" and remain optional. A good option, but an option nonetheless.
Without CPU extensions what you have is emulation (well, binary translation at best) which takes you even further from "baremetal".
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Sure. Because we define commodity hardware as not having that feature. Add that feature, to anything, and we stop calling it commodity.
It isn't like virtualization is unique to commodity hardware. In the past, it was the one type that didn't have virtualization (hence why that market was left with a knowledge cap leading to confusion when it was suddenly introduced to something that the rest of IT had long ago standardized on and accepted.)
I call standard x86 servers commodity. And you?
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Hypervisor types do not make sense (because if you ask a vmware salesman you'll hear the baremetal-not-baremetal argument right there). Nothing works on the hardware directly, everything goes through layers.
They make a lot of sense. The issue is that non-technical salesmen do not make sense. That's a completely different issue.
Something always runs on bare metal. If you use the standard definition of bare metal, then it's a requirement for a computer to work. Bare metal cannot be avoided. The question is only "what" runs on the bare metal.
So this is critical, you can't argue that the definitions of hypervisors are wrong unless you accept the universal definition for bare metal. Because one relies on the other. All hypervisors are defined in relationship to that standard. You can argue all you want that the definition of bare metal is bad, but it is what it is and is universally accepted.
So your issue is not with the definition of the hypervisors, but not liking a part of the definition on which they rely.
Example: you don't like calling blue blue, but call it green because that's your opinion.
Then you claim the US flag is red, white and green. Because to you, it is.
That doesn't change the colours of the flag. So you can't tell someone that uses blue to mean blue that they don't know the flags colours. But if you want to argue that he uses a different word for blue than you use, that's a different thing.
-
@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:
Sure. Because we define commodity hardware as not having that feature. Add that feature, to anything, and we stop calling it commodity.
It isn't like virtualization is unique to commodity hardware. In the past, it was the one type that didn't have virtualization (hence why that market was left with a knowledge cap leading to confusion when it was suddenly introduced to something that the rest of IT had long ago standardized on and accepted.)
I call standard x86 servers commodity. And you?
If you do, then you DO have hardware virtualization options, like I had said previously. But most people only accept certain classes of AMD64 (not x86, that's long dead) as being commodity, not all universally. Hitachi's true hardware virtualization on AMD64 products were not generally considered commodity.
For most people, commodity really means PC architecture, AMD64, without extension hardware upgrades. They dont' write it that way, but anything outside of PC, anything outside of AMD64, or anything with serious hardware enhancements are not lumped in.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Without CPU extensions what you have is emulation (well, binary translation at best) which takes you even further from "baremetal".
That's one option, but not the only one. And if the emulator were to be bare metal, then it's not further from bare metal.
-
@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.
