Submitted by In the view of Mendel Rosenblum, chief scientist and co-founder of virtualization vendor VMware, today’s modern operating system is destined for the dustbin, a scenario unlikely to please Microsoft or any of the Linux vendors. Rosenblum’s keynote on Thursday wrapped up the LinuxWorld conference, preaching the virtues of virtualization, which he believes will eventually make today’s complex, some would say bloated, operating systems obsolete. “It’s just going to go away“, Rosenblum said.
Hype to sell his crap for servers.
http://www.theregister.co.uk/2007/08/09/linuxworld07_ideas_virtual_…
http://www.theunixblog.com/2007/08/virtualization-makes-it-harder-t…
Not only for servers. Most likely that the hype of their coming IPO blowed his mind away…
Did he forgot to take his pills? So what is he going to virtualize if not operating systems??? Air?
Title is misleading, if you read the article, it’s actually the death of general purpose operating systems. He thinks that future operating systems are going to be specialized for single tasks. Something like what rPath is doing.
erm.. Hardware. VMware virtualises the hardware.
An operating system is a must; without it a computer is just a fine piece of plastic, metal, and stripboard.
An OS is a fundamental part of the system. It schedules tasks, manages file structure, storage, provides a environment for applications to execute, mediates hardware and software communication via drivers, and presents a user interface. An OS probably does other things too but I cant think of them all, ATM.
If this guy means a different type of OS… OK then. But its very likely that people will demand even more out of their systems in the future; therefore future OSes will need to be capable of fulfilling these demands.
Edited 2007-08-11 01:18
Time to roll out the network computer, ala Sun.
Yeah, we all see how well that has worked!
It’s very entertaining to take all the various major changes that prognosticators prognosticate and see how much they got right and wrong down the road.
So, I ask you now: where is my flying car? (I know they do exist in some form, yes, but I can’t afford one, and neither can most hope to afford one – assuming you solved all the other issues beyond technical).
Until virtualization software manages everything in a very fine-grained manner in the same way every OS people want to use does, and does it using all the same priorities users desire, virtualization software cannot replace the operating system, and if it did all that, it’d only be yet another operating system, just one specialized for running operating systems.
“It’s very entertaining to take all the various major changes that prognosticators prognosticate and see how much they got right and wrong down the road. ”
Yeah, I’m pretty disappointed about quite a few things:
* Where’s my personal robotic servant?
* It’s pass year 2000 and we’re still here? No cataclysmic nuclear winter yet? I was so looking forward to a dystopian future with rogue cyborgs and radioactive deserts.
* How come we don’t have any colonies on the moon? or at the bottom of the sea for that matter?
* TV? I was promised a holographic projection damnit!
* What is SkyNet up to these days?
http://en.wikipedia.org/wiki/Skynet_%28satellites%29
The way he talks about virtualization, it sounds to me like virtualization is going to be this decade’s internet appliance. (Remember that idea from the late 90s?)
From TFA:
Rosenblum favors a world in which a virtualization layer is tied directly to the microprocessor and other related hardware of a computer. Running on top of this layer would be virtual machines, or mini-operating systems, that would be designed to run specific applications
So essentially he believes everything needs to go back to real/unprotected mode and that every application should introduce it’s own runtime stack from top to bottom on top of a standardized hardware platform provided via the wonders of virtualisation?
Personally I just don’t see it. All he’s talking about here is rearranging various layers of abstraction that already exist today throughout the desktop/server stack.
It seems to me all this would do is push all that nasty complexity from the operating system to application developers where it would result in duplication of functionality in thousands of application tweaked variants. It sounds like Window’s DLL hell all over again! Just consider how totally under utilized 3rd party (non-Microsoft) shared libraries are on Window’s compared to the likes of your average Linux or BSD system.
Programming is all about where you put the fence between you and the hardware. Somehow I don’t think moving the fence back and forth really helps anyone except those trying to snap up property or sell servers on the other side.
Edited 2007-08-10 22:14
You don’t get DLL hell. With current H/W virtualisation, you duplicate each OS / DLL set anyway. Having individual customized libraries actually makes sense in this context. You can really cut out the crap and just bundle what you need for your program.
Sounds like he’s basically talking about an exokernel design. The idea is to have a small kernel that just securely and with good performance divvies up hardware resources. Each application is then tasked with using those resources as it needs to. Now, of course, each application will probably use libraries to do most of the heavy lifting. And if you think about it, a kernel is no more than a library shared amongst all programs (with the added benefit of hardware protection). It’s basically a library OS. I’m sure there can be a way to share the libraries amongst the different tasks running on the system without wasting resources.
Yeah, you’re right about this. Not only is it kind of pointless, but I would take issue with the “more secure” line.
Building something from the ground up isn’t going to make things more secure… the OS disappears and the application is suddenly running with full privileges in its sandbox? Sounds like an easy way to compromise all of the data at once to me.
Virtualization is great, whether it’s hardware, OS-level, user-space-level, or full emulation. They all of their benefits, but claiming the OS suddenly disappears needs a lot more explanation. Not to mention there is free software to accomplish each of these virtualization levels anyway.
I’m surprised vmware is going public right now. They may have a home providing support to some enterprises, but my guess is that their profits will be decreasing as open source alternatives continue to make huge strides.
Application virtualization is the buzzphrase of the month it seems. Everybody is throwing their hat in the ring. But nobody agrees on what application virtualization should be.
I don’t even think it’s a real term. None of these so-called application virtualization solutions actually virtualize applications. They either virtualize hardware or operating systems.
The VMware approach is, of course, based on hardware virtualization. Again, remember the rPath article from not too long ago:
http://www.osnews.com/story.php/18312
This is precisely the VMware model: use a Linux-based package management system to build guest images for a standalone application environment. These images becomes the lingua franca of software distribution, since they are ready to run on any of VMware’s virtualization products.
However, the justification is screwy. If modern operating systems are so complex and bloated, why package each application with its own beastly OS? The OS won’t go away in this model, it would actually multiply! Your computer would run dozens of operating systems concurrently in order to run the same apps you run today on a single OS.
If you want to trim the fat and still provide isolated, manageable application images, OS virtualization is the proper solution. The only downside is that the images must be built and packages for a particular OS (read: distribution and release).
For example, Microsoft’s SoftGrid solution:
http://osnews.com/story.php/18423
Or Zones, OpenVZ, Klik2, etc.
For a corporation that typically standardize and even certify their system images, this requirement isn’t such a big deal. For consumers and small businesses, OS-level application virtualization provides another packaging option for distributions. For example, you can install your webserver environment using APT, or you can just go for the complete LAMP stack as a VPS.
No, operating systems aren’t going anywhere. Of course, neither is hardware virtualization. But it should only be used in situations where multiple OS types, versions, or configurations are required to meet the needs of the applications. Most of the time, OS virtualization is the more efficient option.
Edited 2007-08-10 22:43
You do not need to give such a detailed explanation.
Everyone understands the difference between virtualization software and OS except Mendel Rosenblum .
🙂
You do realize that Mendel Rosenblum isn’t some empty suit? He’s a professor of computer science at Stanford and the author of several dozen scholarly works in fields ranging from filesystems to processor architecture.
The fact that he’s pimping his warez doesn’t mean he’s an idiot.
Am I the only one who takes most of CS professors’ materials as the model not to do something?
Disagree with all of you so far. The article title misleads.
The “operating system” as defined by Microsoft is going to be redefined not killed. VMware’s software is going to control the hardware, not the traditional OS anymore. Which one of these systems you call the “OS” is arbitrary.
The key here is the power of hardware abstraction. This will be a big deal, good for all of computing.
In reality though, this technology is going to be embedded. Think of it as just a major upgrade to BIOS.
Yes, the OS doesn’t manage real hardware in a hardware virtualization setup. But it does manage virtual devices. So instead of targeting a wide variety of hardware, the OS targets something along the lines of QEMU. That’s how KVM works, in fact.
The OS also has to manage its “real” memory, which is really virtual memory in the hypervisor. It has to manage processes and threads, provide file I/O, process interrupts injected from the hypervisor, and implement a variety of system services and libraries.
Hardware virtualization doesn’t change any of this. It only abstracts the bare hardware. So the OS as a hardware abstraction layer fades away, but the OS as a runtime environment remains. It will take a few more years for the runtime space to consolidate to the point where hypervisors can reasonably abstract runtime environments.
As noted in another comment, BEA is introducing a JVM that runs directly on VMware. They claim that they’re getting rid of the OS, but they’ve really just ported a JVM to a minimal OS, probably Linux-based. It doesn’t support all of the features of Java on a real OS, so many existing Java applications will need some porting to run on BEAs new solution.
Edited 2007-08-11 05:22
Mendel Rosenblum is brain damaged by his eagerness for the profit of Vmware.
When debating never criticize the person, always criticize the concept. It makes you look more believable too and helps to preserve a friendly forum for discussion.
Does VMWare have any job openings? Having read the headline, I think I can be a chief scientist there too. What amazes me with all these Virtualization dudes, is that no one has talked to IBM, it appears. They’ve done VM for nearly 40 years, and still have very “complex” OS’s running as guests.
Edited 2007-08-10 23:07
Why I cant stand VMWare.
They, vmware and their related vendors are selling, selling and not innovating . Everyone wants a piece of the VMWare pie but they don’t know why. They have become the GeeWiz word of 2007. One VMware engineer tried to tell me VMWare will do SSI Clustering ( Single System Image ). When we asked how does it share memory or cpu cycles he back peddled and said well it does not do that, but you can fail a vm from one box to another. I asked them a dumb question and got a great response. “Are you planning on making a version of VMWare for Itanium ? ” VMWare sales rep “Why would we want to do that ?” Its Crap to sell servers and nothing more. Long live Jails Zones and Galaxys .
The better architecture is to build software for what Rosenblum called a virtualization appliance. Software makers could package within a virtual machine only those components needed to run a particular application. “I can start simplifying these things,” Rosenblum said. “I can take out parts of the OS I don’t need for this application, and build an OS that’s highly optimized for the application.”
And what happens when these applications need to interact with each other? What’s going to oversee all that? You’d need some underlying framework to ensure that all of these virtual machines understand the various metadata that would be being passed back and forth. In other words, you’d need an OS
It’s called networking
And exactly how is networked drag-and-drop and process interaction less complex that what we have to day? Good thing that doesn’t introduce an absolute shitload of security concerns.
Bah.
“My product, virtualization, is going to rule the future”. Wow, now there is something new…or not. I mean, seiously, who could ever expect a statement like this from a company that *makes* virtualization products?
Well, I’m sure VMWare’s shareholders are pleased with his statemnets, at least.
It hasn’t been a problem for X11, or dcop or anything else. I mean, most of the interprocess communication LOOKS like networking anyways and often uses networking elements like sockets and named pipes.
Fact of the matter is, a traditional OS is already along the lines of a virtual machine manager anyways. Each process is in it’s own sandbox with CPU and memory being virtualized. A virtualization solution like VMWare is nothing more than just virtualizing everything else. It seems like a logical step.
Mr. siride wrote:
[quote]
[/quote]
Actually there is MorphOS.
MorphOS it is a MICROKERNEL based OS and provides HAL capabilities.
Any API of any Operating System can run in separate Sandboxes.
Actually there are only 2 Sandboxes active:
One Called Q-Box (it is the core of the system, and manages low lwvel functions. You may call it like an “Hypervisor” of some sort), and another called A-Box.
A-Box provides AmigaOS 3.1 API and emulation…
—
WARNING: Remember to consider AS-IS this fact regarding Amiga and AmigaOS API!
It is just the first Sandbox made availble to morphOS!
Infacts AmigaOS is just one of the most easyiest OS that could be wrapped into a sandbox.
Actually A-BOX contains the AmigaOS 3.1 API, the device drivers, ELF capabilites to run PPC programs, the Ambient GUI, and an emulator to run 68000 based applications and so-on.
Soon all the device drivers will be moved into Q-Box, and the A-Box will remain still as the first OS available as version available in a sandbox for MorphOS.
Q-BOX will became the complete baseline for any Sandboxed OS.
—
More OS could be sandboxed into MorphOS if you have the capabilities to develop one OS into these Sandboxes.
http://en.wikipedia.org/wiki/MorphOS
Read infos and news at MorphOS sites:
http://www.morphzone.org
Read also about MorphOS into “The pegasos Book”:
http://www.charra.fr/
http://thepegasosbook.charra.fr/
Except….VMWare wants to become THE operating system.
Which will happen.. over my dead body!
Edited 2007-08-10 23:50
“Except….VMWare wants to become THE operating system.”
It seems that VMWare wants to become the OS in terms of controlling the hardware, but all of the services that apps use is still to be provided by the “OS” that VMWare is virtualizing (or the “OS” that is running on top of the hardware that VMWare is virtualizing?). I’m not sure that’s a win.
Anyway, as stated above, if every app is running in its own customized virtual OS, how do they interact? How does something like OLE work? How does something like Copy/Paste or Drag/Drop work? How do pipes work? Seems like a lot of handwaving is going on here, typical of someone from an ivory tower of academia (Rosenblum is a Stanford prof, according to an earlier post).
And the part where Rosenblum talks of removing parts of the virtualized OS from that the app “doesn’t need”, doesn’t simplify things as he claims, it only adds complexity. Because now if a problem arises, you don’t know if the problem arose due to the fact that part of the OS is missing. It’s like going into Windows system directory and removing a dll that you don’t *think* your app needs. Foolhardy.
Edited 2007-08-11 04:09
I agree with this first part, I don’t know how VMware is going to capitalize on the hypervisor. It seems like a powerful position to be in though.
These are all really good questions here. Here goes…
These Virtual Machines will interact with eachother in a variety of ways. There are tradeoffs in how you configure things though, if you want more security, you have to sacrafice convenience and speed.
You will be able to change the security attributes of functions like copy and paste, file transfers, or rDMA. Who do you trust?
. I think in a very reliable setup they are achieving about 90% efficiency on the most demanding computing environments like databases. If they could get a little cooperation from chip design and guest OS optimization I think they can get pretty close to 99% efficiency.
As for your last comment, I don’t think Rosenblum is referring to Windows. I think he is referring to the Linux kernel which is totally customizable. But Microsoft could do it too. In fact, they already are. They have been releasing betas of Windows Server 2008 without GUIs and unnecessary services for a while now. I don’t know about you, but these sound a lot like the virtual appliances based on Linux that are being offered for use on VMware.
VMware IPOs on Tuesday. They are going to skyrocket. But it’s only the beginning. Make sure you all pay attention to VMworld in a month. There are some very heavy keynote speakers scheduled(Intel, AMD, and Cisco). I think that they are going to announce something big.
“VMware IPOs on Tuesday. They are going to skyrocket”
Wow, what a coincidence that he makes this speech just a few day before that, eh?
A company that wants to maximize it’s potential by generating publicity just before it IPOs? Tch, what’s the world coming to?!
to me it sounds like java all over again.
Browser: Mozilla/4.0 (compatible; MSIE 6.0; ; Linux armv5tejl; U) Opera 8.02 [en_US] Maemo browser 0.4.34 N770/SU-18
It probably will end up the new java. Which is a shame. There are 2 reasons that Java isn’t a universal standard today:
1. Intertia. People are afraid of things that they don’t know.
2. Microsoft.
Java may have its many faults. But it is far better for most uses than C++ or C#.
it has also gotten a rep for bloat.
i hear people that dont want to use anything java based because of that…
Yeah, I know what you mean. Although .net and c# isn’t any better
and yet it seems to be used for all kinds of insane things…
But on the other hand, the traditional OSes can as easily also gain virtualization features, rendering VMware useless. And I bet VMware is the one that’s “just going away” faster than Windows or Linux.
Correct, traditional OS’s can, and already have gained virtualization features.
However, VMware is selling two concepts that set it apart.
1. It is going to keep the core “base OS” or hypervisor, very simple and efficient. This is important because you can’t have security holes or anything that could undermine stability in the hypervisor. A hypervisor crash would kill all the guests.
2. It is promising to remain vendor neutral.
These 2 things are much different than Microsoft’s Virtualization environment, but not a lot different than the Virtualization that is available through Linux.
VMware’s main obstacle, as is evident by the comments on this board, is in establishing trust that it will remain vendor neutral.
My bet is that after VMware establishes dominance in this market and taps into new revenue streams, it will move it’s ESX technology to GPL.
No matter how you look at this, it is bad for Microsoft and good for Linux.
Anyone remember thin clients to kill PC’s?
Well… I aint using thing clients here
You guys are sure quick to jump on Mendel and VMware. Lets look at a bit some current events before we judge this guy. FACT: Dell announces that they intend to integrate virtualization into a chip on the motherboard, so all you have to load up is the virtual machine, not an underlying OS. FACT: Parallels on OS X touts its ability to run Windows apps in a stand alone window. FACT: We are already moving services to run on their own virtual machines so that they cant be used to corrupt other services. FACT: EA is moving all their OS X game development to an emulator based environment. FACT: Valve and id just release 24 id games on Steam using DOSBox.
SPECULATION: Is it really that hard to imagine that what we consider an OS is going to be replaced with something that runs in a virtual environment? This guy sounds like he’s right on the mark. Why do we need an entire OS if all we want is to run a browser? Or an ftp server. Why worry about conflicting software when each can be run simultaneously from its own little world on your hard drive? The world is going to virtualization like it or not.
“This guy sounds like he’s right on the mark. ”
No, it sounds like he wants to sell his product. Nothing wrong with that but we should take it for what it is.
“Why do we need an entire OS if all we want is to run a browser?”
Why Do i need a car that can go 100km/h+ is I’m only going to use to and from my job? Why does it need lights if I only intend to drive it in the daytime?
You need a whole OS because otherwise you’ll need to bloat the application with all the hardware drivers and other infrastructure components it needs.
“Why worry about conflicting software when each can be run simultaneously from its own little world on your hard drive? ”
Because if you do this every app need to have their own custom copy of the OS.
Also, how does programs interact? drag-n-drop between different VM’s? How does that make things simpler?
> “This guy sounds like he’s right on the mark. ”
> No, it sounds like he wants to sell his product.
> Nothing wrong with that but we should take it for
> what it is.
Linus and Stallman want to sell Linux. Let’s just ignore everything they say about everything.
> “Why do we need an entire OS if all we want is to run
> a browser?”
> Why Do i need a car that can go 100km/h+ is I’m only
> going to use to and from my job? Why does it need
> lights if I only intend to drive it in the daytime?
> You need a whole OS because otherwise you’ll need to
> bloat the application with all the hardware drivers
> and other infrastructure components it needs.
That’s why we have these things called libraries. The other great thing about virtualization is that you won’t need ten-thousand specialized drivers, you will only need one set…for the virtualized hardware. Apps will have as much of these libraries as needed to do their task, instead of having to have as a requirement this one giant library (the kernel) at all times as it is now in current OS implementations.
> “Why worry about conflicting software when each can
> be run simultaneously from its own little world on
> your hard drive? ”
> Because if you do this every app need to have their
> own custom copy of the OS.
> Also, how does programs interact? drag-n-drop between > different VM’s? How does that make things simpler?
They will interact the same way they do now. Just because they aren’t in a VM now doesn’t mean that they aren’t already sandboxed from each other. They must use IPC which could easily be generalized to a VM situation for a VM based operating system, perhaps built on top of networking.
Read this: http://en.wikipedia.org/wiki/Exokernel
Yet the need for communicating with specializated devices won’t disappear. Perhaps the operating system won’t have to deal with the hardware, but the hypervisor will. Sooner or later, a variety of different hypervisors will appear, each with its own API, requiring drivers written specifically for them. Then we’re back on square one, albeit with a brand new layer of unnecessary software bloat to justify the purchase of new hardware…
The problem, dealing with the complexities of hardware, wasn’t solved. It was just moved.
Ignoring the fact that most applications don’t have to call kernel services directly, what would be the benefits? Right now, I can only see drawbacks. Libraries won’t be shared, wasting resources (unless the hypervisor deals with that, which would bloat it even further). Vulnerabilities in one library would require updates for ALL applications. New communication facilities between applications will need to be needed, forcing developers to reinvent the wheel once again. Etc.
Now, I’m pretty sure that I don’t see the situation like a computer scientist or a software engineer as we have a different vision/experience of computing. Still, writing applications using cross-platform frameworks like .NET or Java seems definitely more efficient to me. There won’t be just one hypervisor.
Mr. Wrawrat wrote:
[quote]
[/quote]
Just read my previous comment…
MorphOS actually deals with A-BOX that requires its own API and its own device drivers (monitor, graphic cards, etc).
Only Q-BOX deals with the hardware (motherboard).
When also device drivers will be embedded into Q-Box or installed and loaded into it, then A-BOX will require only Amiga API, and will deal only with it…
But A-Box it is not the limit, neither it could be the only sandboxed OS available for MorphOS.
Then (for example) we could have also OS-Box (Open Solaris Box) with its own API, but no need for device drivers anymore, or perhaps QNX-Box running QNX API, but again, having devices stored into Q-BOX there will be no need of device drivers written just for QNX and not usable into other Boxes or other OS environments.
MorphOS it is an easy solution. It is available, and it it is a mix of Closed and Open Source.
If only licence fees for Closed Source parts will be reasonable, then anybody could be interested in developing new sandboxes for MorpHOS, hosting new OSes.
Or perhaps, if the owners could be very smart, then they will release the remaining closed parts of MorphOS as Open Sourced.
This will be sweet!
Remember. MorphOS could morph into any shape or include any kind of OS you like to embed into it.
Also, how does programs interact? drag-n-drop between different VM’s? How does that make things simpler?
That’s a beautiful point. After running separate virtualized apps communication will have to exist between the applications for things like drag and drop. Which brings you back to libraries of code shared among applications, which sounds a lot like modern day linux GTK event modules.
http://www.informationweek.com/news/showArticle.jhtml?articleID=201…
Just what we need is another monopoly.
An OS sits between the hardware and the apps. What does VMware do? It sits between the hardware and the apps (OS in this case). In other words, you could call VMware a rudimentary OS. VMware is just another OS, albeit not as bloated as some OSes.
Or did I get it wrong? Isnt VMware just another OS?
How is VMWare going to virtualize all the different capabilities of different hardware?!?
If all hardware types would have the same set of functionality, then it wouldn’t be a problem (also we wouldn’t have 1000 different drivers for lets say NIC’s).
But the world is not as rosy as that…
And just how on earth does VMWare expect to replace the host OS?.
Or as some suggestions go, even the guest OS??
How is it going to supply the necessary API-s and stuff?
Hope some folks are familiar with VM (Virtual Machine) from IBM. The rationale was to implement one operating system that interacts with a hardware and creates multiple virtual machines. Then inside each machine you can run a simple straightforward single user single task operating system (such as CMS). The whole thing came as ‘VM/CMS’. That was the original scenario. The reality was quite different. First, even with such a simple system like CMS there was a substantial virtualization overhead, so from being simple system capable of runing in native mode, CMS became tightly coupled with VM through various hooks. Eventually IBM had to write those hooks in microcode to make the performance passable. Then having VM/CMS, how do you run some multitasking multiuser based application like database. The original solution, inter-virtual-machines communication didn’t work quite well, so for those who wanted to write a software for pure VM/CMS environment, they had to patch CMS to enable it to do multitasking, asynch IO etc (at some point I was working for a software company doing just that). Still that was a hack, and inspite of its popularity in IBM mainframe world, VM/CMS never quite caught up. Now VMWare is claiming exactly the same thing: we need just a virtualization layer, and multiple virtual machines can run a simple OS. Well, seems like history repeats itself. Blah …
Yes, what Mendel Rosenblum is talking about here is a commodity version of IBM’s VM.
VM (or z/VM as I think it’s known now) is IBM’s hypervisor—and like so many pieces of mainframe technology, Mendel Rosenblum is envisaging a scenario where the world of VM (where each user gets his own private virtual machine—or, if the administrator allows—multiple virtual machines) will filter down to the PC world.
There’s a lot of merit to it. Hypervisors are already gaining traction (look at all the buzz around Xen, for example), and EMC/VMWare isn’t exactly without experience in the field. In the mainframe world, it works stunningly well: each virtual machine is, as far as the OS (running on top of the hypervisor) is concerned, a real mainframe. CMS, because of how it’s been developed, was once a standalone operating system in its own right, but has long since been developed in tandem with VM, but there’s nothing about VM that means you have to use CMS: IBM push Linux under z/VM quite heavily as a platform.
The communications aspect is certainly one that need consideration. IBM got around it by creating a special inter-VM networking layer called “hypersockets”, which is basically a clever virtual networking environment tuned for high-performance transfers between individual virtual machines. VMware, and friends, already have the ability to provide virtual NICs as it is—if they thought there was demand for a tuned variant that had reduced overheads (e.g., by using buffers that were shared between the VMs), then it’d be implemented… and it probably will be before too long if things go in this direction.
The key thing is management, though. With a hypervisor like VM, you can decide exactly how much in the way of resources should be allocated to each VM’s operating system (or “guest” in PC virtualisation parlance), and adjust it on the fly to suit workloads. VMware, et al, are really only just starting to get these kind of features.
In the mainframe world, of course, there is no “host” OS: the hypervisor is the host, and it provides its own interface to users (including account management, networking, and so on). This has benefits in that you don’t need to worry about—for example—XP sucking up resources when you want to run a Linux guest, or vice versa. In the z/VM scenario, the hypervisor’s relatively lightweight, so it doesn’t impact (in a negative way) upon the VMs.
Translate this into the PC space, and you’re looking at a scenario where you boot into the virtualisation product (VMware, Xen, whatever), and it presents you with a login, and opens network ports for remote logins, depending on how it’s configured. Presumably this would have to include the likes of VNC and MS RDP to cope with graphical OSes. Once you log in, you start up your guest OS, and it runs. If you’re the only user on the machine, and you’re only using one VM, your guest OS has roughly the same performance you’d expect from running it on bare hardware.
Really, this is somewhere that paravirtualisation products have been headed for a while. I think it’ll be a little while before we get there—building a lightweight but networked secure paravirtualising hypervisor isn’t exactly the simplest thing in the world—but it’ll be cool when it happens. I don’t think anybody believes that VMWare will be the only product actually going this way, though.
You don’t require to run CMS under VM – yes. IBM runs multiple Linux virtual machines under z/VM to show that having a big iron with multiple virtual machines offers substantial advantages vs having the bunch of PCs each runing its own version of Linux. Advantages such that management and networking. Agree with that. But what VmWare proposes is exactly runing some scaled down operating system under hypervisor and I”m saying that will be akin runing CMS/VM, and that’s just not going to work.
As I was working with VM (VM/XA and VM/ESA but not Z/VM), I ran whatever can run. All flavors of MVS and even VM itself (‘second level hypervisor’). Great for development shop and for testing the new release of OS.
In any case, take everything coming from VmWare with a grain of salt. They are on a verge of IPO, ‘ya know.
I don’t want to sound like I’m trolling, still I’m very skeptical. All this virtualization business seems to be another case of “let’s add another layer to our computing environment because it’s too complex.” Way to go!
While virtualization may be the wave of the future, current operating systems will just integrate their approach into their current product. Comming up with a stripped-down version that just runs the “virtualised applications” doesn’t seem to far-fetched either.
Linux has already jumped on the train, and as soon as it becomes even more popular, you can bet Microsoft will follow as well (not just virtual PC – something integrated in Windows) – which means that *if* anything is about to die, it’s vmware, inc.
Microsoft will follow as well (not just virtual PC – something integrated in Windows) – which means that *if* anything is about to die, it’s vmware, inc.
Looks like MS is almost there. See link.
http://apcmag.com/6891/microsoft_softgrid_taking_hold_of_virtualisa…
Soon the Rebol language by Carl Sassenrath will enter the stage of being the first computer programming language which will be capable to boot by itself, and provide its applications to run on any OS, or even run LIVE because each application will provide its own hardware abstraction layer, I/O data and database access, a standardized GUI, multimedia capabilities on the fly, access to storage media on the fly, networking services and capabilities of inter-applications communications and data parsing.
Rebol applications will do that even in a VMWare environment or either into a LAN or a WAN like Internet is.
At least this is the final target of REBOL Internet OS as intended by Carl Sassenrath!
WOW!
Will Rebol became the new language for VMWare based environments?
http://www.rebol.com/
http://www.rebol.com/index-ios.html
What is Rebol IOS (Internet OS)?
http://www.rebol.com/ios-intro.html
Edited 2007-08-12 15:20
Essentially, VMWare’s rep is hand-waving exactly what these “mini-OSes” would comprise and look like — and then what the implications would be. Don’t get me wrong: I’m not opposed to looking at a potential alternative to today’s operating systems; however, it’s my opinion that we would end up replacing today’s OS complexity with a future blend of app complexity. So, what exactly have we gained, other than pushing what was previously operating system functionality higher in the stack — up to the application space? App writers have more than enough on their plates without worrying about without having to duplicate or supplement OS functionality. Not to mention, will app writers have the expertise to even accomplish it? I’m not so sure that they would be willing to sign on to do this work. What’s their incentive? To elimiinate today’s operating system? Why would they do that? They get plenty of value out of Linux and Windows and Solaris and AIX. This seems like yet another “solution” looking for a problem to solve.
I very well could see virtualization becoming the base for every operating system. Stuff like the current “shared hosting” will die and be replaced by each user getting their own virtualized hardware, based from an original chip inside the server (ala Dell) so users won’t affect each other on shared hosting. The same can be true on operating systems to split between the users as well. Instead of the operating systems trying to be multiuser, which has proven to be troublesome for simple minded people, each user can have their own virtualized operating system so they really can’t affect other users at all on the system. Virtualization definetly has a place in the future, although I think he is being a bit extreme.