There isn't much to go on here, and a search for +Mendel +Rosenblum "virtual client" doesn't reveal much. Did that man just say "remove the OS and replace it with VMware"?

Well - if you really wanted to, you can do it now. Strip down your Xen-based, paravirtualised FreeBSD or Linux kernel so it's just a scheduler, a memory mapper and a filesystem. Virtualize your storage with a whatever appliance or SAN, and just boot Oracle over that very thin layer.

Okay, you could make a whole new lightweight framework for plugging in just those things. And of course Andy Tanenbaum will love this notion. And of course we're practically there already. Oracle plug their ocfs2 into Linux quite nicely.

The trouble with the minimalist approach starts when you want to manage needs that accumulate in production. Suddenly you start to need filesystem management tools, debuggers, profilers, scripting languages, SNMP services, compilers, test harnesses, firewalls, volume management, development tools and so forth. And before you know it you've got a general-purpose OS. And it's not so bad, because you're getting work done. But why reinvent that wheel? For each new application? Sounds like a an exercise in accumulating multiple, inconsistent system management models like moss.

I realise that being a Professor doesn't have anything to do with the dirty intricacies of managing real-world production systems. In this case it sounds like an compulsion to creating the cleanest possible software stack, hang the consequences and realities. Which I'm sure we all recognise as a classic nerd urge. And I'm sure every manager recognises as a problem, not a solution.

Still, his point seems moot. With good packaging models you *can* strip it an existing system down almost trivially. And if you have a competent sysadmin they can do this now with Solaris, BSD, Linux - at both userland and kernel level. And with a braver admin, with Win2003.

though doing this in VMs rather than trying to pimp hardware does seem like a better business model from the customer's perspective

...but if you had a virtualware application that could run multiple dedicated "virtual OS's", dedicated to different applications... wouldn't that be the same thing as an Operating System??

They said the same thing about how linux was going to revolutionise the datacenter... still waiting...

You said it all. [Y]

In the 'good old days' when there were multitasking OS's that would take up just 48kb (Sinclair QDOS\Minerva), a programmer could write an application that would be able to handle all the channels & ports required. Now with bloated OS's it has become more difficult to write small modules of code to do the same thing, without ending up with a bloated programme.

Tell you what, let's take this minimalist kernel, and call it, for example a "microkernel" with just enough features to support the database and some basic drivers. Then, as the first comment pointed out, there will be routine housekeeping stuff to do, which will need (e.g.) file systems and file managers and maybe an editor or two, perhaps a compiler and an IDE and some backup tools, not to mention some non-core drivers and stuff, maybe a windowing system for GUIs... If we pick the best of those on the market together with the best microkernel on the market, what do we have?

We have the Open Software Foundation's "best of breed" process that brought us the OSF/1 operating system and OSF/Motif windowing and ... oh, but that was before enterprise-ready Windows and long before Linux. Things would be so much better in the future if everybody just used VMware instead of the OS. Not.

Call it what you like, there's still a need for an OS. He might argue that there's no need for device management, but at some level there has to be. A virtualized system still needs something to run on. He's selling a virtualization product, of course he'll say what he's going to say. I think everybody should have a full blown security assessment once a year. Guess what I sell.

Let me see.....software that's used to mediate between a number of separate, concurrently running applications, managing resources common to those applications and communication among them and between them and the outside. I'm sure I've heard of such a piece of software before.....yes, its an "operating system".

I've been ground down trying to explain to neophytes (and it look like most of the marketing / suits) that a bloated graphical desktop is not an operating system, its a frigging application. But, simple as the notion is, its still seems to be too deep a concept for many.

All VMWare does is put a micro-kernel under commercial operating environments. Given the state of some of the PoSes that pass for "operating environments" its a good idea to keep them on a short leash (and away from the hardware).

He is right about operating systems being crap. You can't just take an application, it's data and configuration and plonk it on a machine and say run here, then pick it up and plonk it on another machine to run tomorrow.

That's why virtualisation is so attractive you can pick up a virtual machine with its operating system, application, data and all the configuration and crap and plonk it on this or that physical machine to run.

Of course an operating system can be made meaner and leaner if it only has one application to run which would save on the horrid overhead of multiple bloated operating systems in multiple virtual machines on the same physical machine.

However, if he is advocating writing or re-writing operating systems then why not write one properly in the first place so you can easily move complete configured working applications between machines and do away with the problems which make virtualisation attractive?

VMware's products are there to work around problems with OSes and applications. If you are going to start re-writing then maybe you should be trying to fix the problems not to make the work around more efficient.

Imagine that, an operating system whose sole purpose is to allow applications to pretend they're sitting directly on top of the bare metal. Why, a system to enable virtual machines should have a snappy name, like Virtual Memory System (VMS for short).

And people say that the only thing reinvented (badly) nowadays is UNIX...

So I've got my my word processor running on a virtual machine, and my web browser running safely tucked away on another virtual machine.

I'm a total hack, so for the paper I'm writing in my word processor, I'm citing a wikipedia article. I go to the web browser vm, copy a snippet of text. Go to my email client on that machine, and send it to the email address on my word processing virtual machine. Then, I switch back to the word processing machine, check my email (its lightning quick thanks to my 8 core processor), copy the annotation out, and paste into my word document.

But its secure!

Isn't a virtual machine an abstraction implemented by an operating system (of some sort)?

Microsoft Windows is like an expensive ill-behaved pet that craps all over everything to mark it's territory. VMware is like a cage with a nice smelling kitty litter under it that keeps this behavior in check. Although VMware makes Windows more managable, it would still be better if Windows could learn it's proper place in the domicile, since keeping it in the VMware cage really isn't what anyone really wants.

Essentially VMWare (or a derivative or a system that preceded it like VMS) becomes the operating system.

I don't think Rosenblum was advocating that we do this tonight while all our users are home tucked up in bed. He's looking to the future and advocating a path that is well worth considering.

Unless we at least start thinking about stripping away all the dross that comes with an operating system that is designed to be "all things to all people", (jack of all trades, master of none), then we are missing an opportunity to innovate and create computing environments that are really stable and do deliver what the user expects.

I like the idea of paring back what gets installed so that only what is needed is there. All those commentators who started listing off what is needed in addition were missing that point - those are applications/features that would naturally be part of the ecosystem that the database or mail appliance or whatever is a part of. You don't need an operating system that is capable of being a dns server or an ftp server when what you really want is an ecosystem that is capable of presenting data to users.

While it might not be novel thinking, it is worthwhile thinking along the lines advocated. Who knows, someone reading this might come up with the universal kernel environment and some custom ecosystems that will eclipse the current status quo - maybe not tonight, but in 10 years.

Rosenblum's quest might be answered by porting the RSX-11D, or it's successor IAS, operating system to current hardware.

It provided the bare basics. It handled interrupts, hardware or software; scheduled processes; and provided interprocess communications.

All devices were handled by a special class of processes known as device handlers. They were special in the sense that they serviced interrupts in kernel mode but otherwise ran as any other applications but at higher priorities.

Except for the physical aspects of I/O provided by device handlers; file systems, databases, keyboards, displays, etc. were applications called by other applications to perform some function.

RSX-11D was implemented by Digital Equipment Corporation in the Seventies. Carnegie Mellon University re-introduced this approach in the Nineties calling it the Mach kernel. Rosenblum seems to want to re-introduce the same approach calling it ... VMWare?

Many years ago, Teradata built DBMS platforms that were stripped to the bare essentials to support their DBMS. Custom "OS" and a single application.

About 7 years ago the concept was ported onto UNIX: the UNIX kernel and basic scheduler and I/O and a "virtual" hypervisor loaded on top of it that allowed virtual machines to run amok on top of the host OS. Today this has moved onto Linux, and does essentially all that Rosenblum ascribes to the hypothetical "thin server". It works well enough to allow complex SQL to process millions of rows in 4,5,6 or more way joins in a few seconds - on paltry dual CPU Xeon boxes at the low end and huge MPP clusters with over one hundred CPS (and a thousand virtual processors) on the high end.

Yes, it 'tain't cheap - but if you're looking for a balls-out database demon this is where you end up.

Rosenblum's vision fits into the concept of "utility computing": each of the "virtual" applications runs on an "expensive" purposed host, but the cost is shared among thousands or millions of users. Picture Google running farms of purposed servers for each of their desktop applications for their future virtual office suite. Works for me.

And speaking of things like clipboard and other functions: there are several models that come to mind for handling this type of functionality, from "local" functions that use the primitives in the display station to handle data exchange to specialized protocols for moving the data from one application to another.

Now, if only someone can fix the internet to support this like electricity...Al, you done messing around with the environment yet?

So how long before we see apps running directly on VMware/XEN ?

If, as many people have commented, the virtualisation system starts to turn into your mini O/S, why not just completely skip the "main" O/S ?

And will we start to see virus checkers at the virtualisation layer as well ?

Someone just used the words 'Windows' and 'operating system' in the same sentence, where one was the subject and the other the object.

Kids!

I agree that operating systems should not stay married to a particular computer.

From this believe G.ho.st started to the revolution in OSs by developing the Global Hosted Operating SysTem, taking full advantage of the Internet to move the entire OS onto the Web. G.ho.st aggregates all of the new Web-based software applications into a Virtual Computer, by providing a Web desktop, a single online file system, a single login, sharing and collaboration. For anyone who wants their personal computing environment to be mobile the G.ho.st Virtual Computer (VC) provides the first real alternative to Windows – a complete personal computing environment accessible from any browser.

G.ho.st alpha version is live today at http://G.ho.st.

Thanks

So:

- My Java App runs in a virtual machine (to abstract the O/S from the App)

- The O/Ss job is to abstract the hardware to a common standard for Apps

- The O/S now runs in a virtual environment (To abstract the hardware from the O/S)

In a few years time, someone is going to write an application that runs directly on the hardware, and they'll discover that we don't actually need all the CPU & memory we've currently got.

First, what Josh said. Although I'd go further : the issue is not with the OS, it's with the applications, and their support. You can strip down Un*x and Windows, but nine times out of ten the application developers either won't tell you all the components their application needs, or don't actually know, or won't support its operation under anything other than the full OS. If I developed an app, I'd want any problems replicated under a full rather than cutdown OS; as the app user do you want the hassle of maintaining a production, and a fully featured problem replication environment?

Second, virtualisation hardware virtualises and abstracts a particular set of hardware. The advantage of this is that a virtual machine may be moved from one host to another (sometimes in real time) regardless of the underlying hardware.

Unfortunately, if an application requires hardware that's anything even remotely unusual i.e. not a network adapter, it's not going to work under virtualisation. Virtualisation is powerful stuff, but it's not a panacea.

Run your database server on a virtual machine...?

It was called Pick.

To a degree this can already be done.

All the virtualisation products that have been discussed here are hardware based / hypervisor products.

What about Virtuozzo (for windows) or OpenVZ for linux?

They virtualise the operating system. You only install one OS (on the host) then virtualise the OS components that are required for the virtual environment.

http://www.swsoft.com/en/virtuozzo/

http://openvz.org/

Just my 2p worth!

[quote]G.ho.st alpha version is live today at http://G.ho.st.[/quote]

Blackhats - Start Your Beanies!

Cue sounds of evil chuckles and hands rubbing together in glee...

I just attended a VMware conference and I think was is missing from the article is the underlying VMware agenda, namely their Virtual Appliance Marketplace. Not that having an agenda is a bad thing...

The basic plan is to do away with the various pieces of black box hardware appliances and replace them with a large(r) standard x86 server running VMware (or any other platform that supports their "open virtualization format"). You do away with potentially underutilized hardware appliances, much in the same way they advocate doing away with underutilized x86 servers. And they do have some interesting appliances out there that have been very effective at removing the general purpose OS and replacing it with just enough OS to do what the application needs (check out the BEA JRockit VM).

Basically I've been working on a java based exokernel system that is running on the java virtual machine.

Essentially, you can use my GUI and libraries or you can develop your own.

It also has an extendable command set, so if you want to roll your own custom command for the system, you can just program it in and add it to the lot.

This gets around what Peter Kay describes as the cut-down problem, if it's in java, then you just run once on pretty much anywhere.

Secondly, apps run in their own library operating system, (you can use parts of the default system to make an app run in its sandbox) so apps supply their own minimums to run.

Only problem I'm having is getting the database system (a part of the filesystem) to learn.

Someone asked what the point is, and I think it's fair to say that runnign Windows on WMware does make it a smoother experience. It boots faster, and with generic hardware all you can see, it tends to have less outages per windows instance right away. But as someone rightly commented, sometimes you still need a physical server for specific apps or dongles! Another nice feature is that you can P2V any windows server - taking the physical and quickly create it as a virtual server. These tend to have odd legacy errors and the like though, as the old hardware management stuff can leave their hooks in (services or unexpected reg entries). Nothing seriosu but an example of the difference that natively building on the ESX platform gives.

Someone asked isn't VMware an OS? Well it is after a fashion - it's a cut down purpose built Red Hat install as I recall (I could be worng, I'm not our VMware guy in here), that provides the OS' it hosts with no actual hardware visibility, just genericised drivers. It also runs it's own management software allowing VirtualCenter in to manage each OS.

Our experience of it has been very postive - it's slowed our physical server expansion radically, and in a few short years we now have hundreds of VMware virtual servers running on our ESX hosts, and it stopped our physical boxes crossing into three figures! So you're saving on space and power right off the bat. It's also handy with P2V, as you can scrap old hardware and keep the server itself in virtualised form. We have had one significant outage when an ESX box went down, taking production boxes on it with the, but comparing the uptime on VMware, it's still a big improvement, and that was the only hit we took - it does highlight hough that you have to plan where you host things, as you don't want all your eggs in one basket.

But we're a large organisation, and we have a good budget - it's a big spend, so smaller enterprises might find a different experience, as

* less support staff means no dedicated support, just extra work for the server admins

* less resources may mean more load on each ESX (we keep production ones hosting no more than twenty or so VMs I think)

* harder to justify the spend of you have a small network, and if you have a low server count, then it might not make ANY sense!

Just my two cents

MikeC

I agree with the users that claim that the OS is not going to disappear ..... it's just being "re-worked" if you will.

I guess that the main point is that yes the VMware suite of products (not only the hypervisor but all the components that build the Virtual Infrastructure) are going to become the Datacenter OS foundation and what remains in the virtual machine is just a very thin layer that the ISV would use to wrap their applications with.

The key point is that this "datacenter os" will be very focused on datacenter needs ...... not general purpose. Why would you want to use an OS designed to support Windows Media Player to run your SQL database in the back after all ?

I also agree this won't happen over night (assuming it will happen of course).

I posted a similar comment on the topic here if interested:

http://it20.info/blogs/main/archive/2007/03/27/5.aspx

Massimo.