What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Actually, a hypervisor in it's cleanest form is exactly that - a driver made to address the extra CPU features that allow for proper virtualization instead of tricks with optimized emulation, like VMWare did before VT. But nobody uses that word in that sense, because if we did - the only actual hypervisor out there would be KVM, all others are a stack of driver plus emulators, plus hardware drivers, schedulers and userspace tools.
Not at all. Hypervisors pre-date those features. It's redefining hypervisor after the fact that makes that seem plausible. But it's not. Hypervisors are not drivers, and hypervisors don't require CPU extensions. That's a very recent thing considering we've had hypervisors for many decades, but the extensions only for about fifteen years.
If you were in the virtualization and hypervisor world before that stuff existed, it would make it obvious how silly it is to say that hypervisors require something that is decades younger than they are.
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
It's all semantics. Semantics is just another term for accuracy. A kernel plus tools for a function has never been the description of an OS. An OS has to be a specific general purpose platform for applications. If you narrow the use, it has to stop being an OS.
That is extremely arguable. Actually, my wife, a VHDL/VLSI/FPGA developer, would laugh if she heard that
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
API on the baremetal. Really. Can you show me the chip that API is implemented in?
Ah, I think the issue is that you don't know how bare metal is used in IT terms. Bare metal never means running on in the chip, it's the code that is run directly on the chip rather than being virtualized a layer up. If you were thinking that bare metal means in the chip's code itself, that would explain the misunderstanding of how hypervisors work.
That term does not mean that. There are things that run on the chip. And there are hypervisors that run in the hardware. Those are called hardware, not bare metal.
-
@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 all semantics. Semantics is just another term for accuracy. A kernel plus tools for a function has never been the description of an OS. An OS has to be a specific general purpose platform for applications. If you narrow the use, it has to stop being an OS.
That is extremely arguable. Actually, my wife, a VHDL/VLSI/FPGA developer, would laugh if she heard that
Well, it's not very arguable. It's the accepted definition for decades. "An operating system (OS) is system software that manages computer hardware and software resources and provides common services for computer programs. "
New definitions to old, existing things are never right. That's "playing semantics". These are well established, well understood, universal terms. You can't just change them suddenly.
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Not at all. Hypervisors pre-date those features. It's redefining hypervisor after the fact that makes that seem plausible. But it's not. Hypervisors are not drivers, and hypervisors don't require CPU extensions. That's a very recent thing considering we've had hypervisors for many decades, but the extensions only for about fifteen years.
OK, replace that with "modern hypervisor". I'm sure you know 15 years in IT is quite the eternity
If you were in the virtualization and hypervisor world before that stuff existed, it would make it obvious how silly it is to say that hypervisors require something that is decades younger than they are.
I was not, but I specialized in the modern version of virtualization for the past 10 years or so
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Well, it's not very arguable. It's the accepted definition for decades. "An operating system (OS) is system software that manages computer hardware and software resources and provides common services for computer programs. "
I see no negation to needing to narrow down the functionality of the OS here.
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
You can't just change them suddenly.
Actually, you can, if you get most other to agree. That is part of how languages evolve and change over time.
-
@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:
Not at all. Hypervisors pre-date those features. It's redefining hypervisor after the fact that makes that seem plausible. But it's not. Hypervisors are not drivers, and hypervisors don't require CPU extensions. That's a very recent thing considering we've had hypervisors for many decades, but the extensions only for about fifteen years.
OK, replace that with "modern hypervisor". I'm sure you know 15 years in IT is quite the eternity
Okay, but now you are excluding other modern hypervisors. You have picked a subset of hypervisors particular to your use case that doesn't include Xen even now, and ignores the hypervisors from vendors like Oracle and IBM today. It's not modern vs. old, it's "things you are thinking of versus ones you aren't thinking about".
You can say "some hypervisors today" to things a certain way. That's fine. But are trying to redefine what has been there in the past, based on misconceptions about a subset of things today.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
I was not, but I specialized in the modern version of virtualization for the past 10 years or so
In a modern form of virtualization. It's still just one of many kinds. And I think that people in the Type-C space would argue that it's not that modern relatively speaking.
The virtualization space, when Intel and AMD started their extensions, and virtualization hit the SMB created a huge explosion in misunderstandings and people trying to reuse existing, very well defined terms for things that they were seeing but not understanding.
Virtualization proved to be beyond comprehension for a lot of IT, so misuse of terms was rampant. For example, most IT pros, even those in the Windows space, actually believe that Hyper-V is loaded on top of Windows and don't realize it is bare metal. Even explaining how it works, or showing that the load method is decades old and very standard, it's still super confusing. Or explaining that things like RDP is not a form of virtualization - because they RDP into virtual servers so the two must be the same thing, right?
So terms have gotten really murky when only looking at the most recent years where we've been immersed in misinformation. But these are all things that have been around a long time.
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Okay, but now you are excluding other modern hypervisors. You have picked a subset of hypervisors particular to your use case that doesn't include Xen even now, and ignores the hypervisors from vendors like Oracle and IBM today. It's not modern vs. old, it's "things you are thinking of versus ones you aren't thinking about".
Wrong. Xen is loading it's own stripped down kernel with a minimal set of features, aka domU (yes, a tiny little OS) which loads up dom0 to manage it all.
Vbox (if that is what you meant) uses an OS and actually has it's own VT drivers nowadays.
All I'm saying is that there is no real true baremetal, it's just not possible unless the stack is implemented in hardware, which isn't the case on commodity hardware.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
The only real baremetal VMs were on mainframes (and I suppose there might be some exotic FPGA implementations), but we're not discussing those right now.
Was on a call, sorry for the delay. But this is the same as I said before... this is called "hardware virtualization". Not bare metal. And it isn't just mainframes, it's available on every class of hardware, including PC. Very rare on PC and totally ubiquitous no mainframes and minis, but every class has had it for a while now. PCs, like always, were the last to get it, but they have it (Hitachi provides this, I believe.)
So this is a key understanding to what we are saying.
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.
-
@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:
Okay, but now you are excluding other modern hypervisors. You have picked a subset of hypervisors particular to your use case that doesn't include Xen even now, and ignores the hypervisors from vendors like Oracle and IBM today. It's not modern vs. old, it's "things you are thinking of versus ones you aren't thinking about".
Wrong. Xen is loading it's own stripped down kernel with a minimal set of features, aka domU (yes, a tiny little OS) which loads up dom0 to manage it all.
How is that not what I've been saying?
But Xen doesn't have a "stripped down" kernel. Xen IS a type 1 hypervisor. That means it IS a kernel (Type 1 Hypervisor = kernel running on bare metal), not an little OS. It doesn't do enough to be considered an OS by any reasonable standard, no one uses OS in that way. If you did that, SNES video games were operating systems, obviously they were not.
That Dom0 provides some tooling doesn't change that Xen itself is the kernel on the bare metal.
-
@scottalanmiller said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Virtualization proved to be beyond comprehension for a lot of IT, so misuse of terms was rampant. For example, most IT pros, even those in the Windows space, actually believe that Hyper-V is loaded on top of Windows and don't realize it is bare metal. Even explaining how it works, or showing that the load method is decades old and very standard, it's still super confusing. Or explaining that things like RDP is not a form of virtualization - because they RDP into virtual servers so the two must be the same thing, right?
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. Now hyper-v works just like Xen. In fact, VMWare also rewrote their stuff to the same model more or less, when they moved from ESX to ESXi.
So terms have gotten really murky when only looking at the most recent years where we've been immersed in misinformation. But these are all things that have been around a long time.
Terms in IT are very fluid, true. This is exactly why I don't like the hypervisor types distinction, and this "baremetal" nonsense
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
All I'm saying is that there is no real true baremetal, it's just not possible unless the stack is implemented in hardware, which isn't the case on commodity hardware.
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.)
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.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Terms in IT are very fluid, true. This is exactly why I don't like the hypervisor types distinction, and this "baremetal" nonsense
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.
-
@dyasny said in What would your recommendation be for a Type 1 Hypervisor - including backup and restoration options:
Now hyper-v works just like Xen. In fact, VMWare also rewrote their stuff to the same model more or less, when they moved from ESX to ESXi.
Opposite. ESX worked like Xen. ESXi does not. There is no "Dom0" at all.
-
@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.
