Vista the Last of Its Kind

An anonymous reader wrote to mention a TechWorld story about Windows Vista. According to the Gartner Group, Windows Vista is likely to be the last of its kind. "The problem is that the operating system's increasing complexity is making it ever more difficult for enterprises to implement migrations, and impossible for Microsoft to release regular updates. This, in turn, stands in the way of Microsoft's efforts to push companies to subscription licensing. The answer, according to Gartner, is virtualization, which is built into newer chips from Intel and AMD, and has become mainstream for x86 servers through the efforts of VMware." Speaking of Vista, C|Net reports that a new release candidate is on the way. The average tester should expect it by the end of September.

Latest BS from Gartner

[owlstead]

And it's late as well. I don't believe so much in virtualization on this level to create security. I mean, how are the components going to communicate? Sockets? Sockets are their own security problem. Microsoft clearly thinks in the same direction. What we need is a more fine-grained security model, in which applications only get the resources they need. Applications themselves also must be able apply the same security directives to their internal components. Just assigning a user per application won't work either, I mean, I would like to continue to work with my text editor as myself.

Currently, applications can install themselves anywhere they want. They can destroy everything I own, including most things in the registry. They can take every bit of CPU power they like. Any amount of memory. Any amount of threads. Any amount of desktop space (including the whole lot through DirectX). They can even take away my keyboard. I don't think you can solve this by just giving every application it's own CPU and operating system. You can do this by restricting access, and by letting the OS take care of the installation and access conditions (maybe not configuration).

The way to do this is to create dependencies with modules, and create security managers to handle access. This is e.g. part of the Java security model, which is sadly hardly ever used. Microsoft has it's own copy of that of course. It's in .net and works with assemblies. Where Microsoft has an advantage is that it owns the Windows operating system, and can therefore easily use a centralized Virtual Machine (as in MSIL virtual machine), installation procedures etc.

I've little doubt that this is the direction Microsoft is thinking for the long run. Unfortunately they don't seem to grasp it on the same abstraction level that Sun can, so it will probably take time. No doubt it will take double that time for Gartner to understand it. Just running every app in its own OS is much easier to grasp, especially when it is already there.

Re:And Linux ?

[broeman]

pornotube.com (don't ask where I know it from ;)

and it is NOT WORK SAFE, that's why I didn't link it. (I wish there were more of these tags on google video, youtube, break.com ...).

Re:Good point.

[It'sYerMam]

More than this, what would be good for the toolkits, i.e. GTK, QT and so on, would be an abstraction layer, whereby the program would load a generic "interface" library, and would make generic calls to it, and the user could select between toolkits, but retain their programs. There would be interesting problems to overcome when one toolkit offers a function that another doesn't, but I'm sure there's a way around it :)

Re:Step 6 of the Software Life Cycle: Death

[mickwd]

"Even bitter enemies mourn the loss of their rivals."

Not in business where it's all about making more money.

Re:MS Windows != Every OS

[senatorpjt]

there's nothing wrong with shipping a kernel and a firewall on the same disk, but the firewall shouldn't be in the kernel!

The firewall is in the linux kernel.

Re:MS Windows != Every OS

[abigor]

iptables is a part of the Linux kernel. There is a userspace program called 'iptables' that acts as a front end, but all packet filtering, rule traversing, etc. happens in kernel space.

Linux does not have well-defined interfaces, particularly not at the kernel level. Just ask any driver writer.

Re:MS Windows != Every OS

[SanityInAnarchy]

Linux has no concept of drivers or kernel modules.

...WHA? Yes, there is a concept of a kernel module.

It's all slapped inside one monolithin kernel, so if you need it to support something else you need to put it in the kernel and recompile it.

I guess that's why I have to download source code to the nVidia driver, patch my kernel, and recompile to make my video card work. Oh wait -- I don't. The nVidia driver gets compiled separately, and produces a module that is loaded on demand. If I don't want to start a GUI, I don't have to load the nVidia module.

And that's only because I compile my own kernel. On binary distros, like Ubuntu, most drivers are included in the core distro, but there are still plenty of add-on drivers you can install as kernel modules in completely separate packages. If nVidia licensing would allow it, Ubuntu could ship a package with the nVidia module, without forcing users to compile it for themselves.

On Windows, if I don't want to start a GUI, I'm SOL. Hell, if I don't want IE, I'm SOL.

As a difference Windows is more of a microkernel architecture (it's in fact a hybrid), where you have many standalone, manageable kernel pieces that communicate in a well defined interface.

No. OS X is a hybrid, at least until they decide to nuke the microkernel parts to get a performance gain. Windows is just a monolithic kernel with enough well-defined interfaces that you can easily ship binary drivers that don't screw up the system. Want to prove it's monolithic? Any kind of driver, if it's poorly written, can crash your whole system -- just like on Linux.

And a kernel no longer defines the OS. Windows is screwed up because Microsoft integrates tons of stuff in userland that have no business being integrated. This is not the only reason Windows is screwed up, but it is a legitimate one.

Re:Last of its kind? I hope so...

[1u3hr]

Where's the 3rd party product that implements a database-like file system with tagging

BFs (BeOS).

Re:By what means?! (and spoiler

[modeless]

Wow, it's obvious you didn't read the article very closely. The 50-layer thing has nothing to do with virtualization; that's Microsoft's *current* (failing) effort to rescue the Windows codebase from collapse under its own weight, without resorting to virtualization.

What Gartner is saying (and they didn't make this up but are parroting smarter people who suggested it first) is that Microsoft should ditch the backwards compatibility that is hobbling them and start over from scratch with a new OS. (I'd prefer something based on Singularity [msdn.com] but that's a long shot.) Virtualization then provides the means for backwards compatibility; simply virtualize Vista in the background to run all your legacy apps and drivers. But this virtualization doesn't have to work like VMWare or Virtual PC today; Microsoft can produce a version of Vista that integrates seamlessly with the host OS when run under virtualization. This is what they're talking about with "integrating data across partitions" and "creating a consistent user experience".

Microsoft has resisted this path, claiming that it's impossible. But Apple has already proved it can work, with Mac OS Classic in OS X. That's what we're talking about here: Microsoft needs to pull an OS X and rewrite their OS from the ground up, then produce "Vista classic" for backward compatibility.

Re:MS Windows != Every OS

[kimvette]

Can you point out the built-in to the OS equivalent of IE in Linux? You can't, because it doesn't exist.

Sure can. Check out KDE's konqueror and KHTML. It integrates very nicely into kwin (the window manager), is embeddable, providing the help engine, integrates with various services/daemons via plugins and extensions, provides integrated and seamless web, help, and file browsing, can also act as a media player (through plugins and extensions), a smarter "search/find" feature than Microsoft will ever deliver in Windows, and if elevated permissions are required for a task, you will either be prompted for the password or the action is simply disallowed (depending on the plugin or extension) unless you specifically start konqueror with su/kdesu. Also, an integrated terminal window is provided so you can run tasks on files in a directory without having to clutter your desktop with separate Xterm windows.

In summary: konqueror in KDE is what MSIE/Explorer tried to be in Windows, and more.

Re:MS Windows != Every OS

[ultranova]

Microsoft, Apple, Ubuntu, etc are all doing this because *that's what the majority of their customers want*.

Dunno about Apple, but Ubuntu is based on Debian, so it has the concept of "packages". This allows the situation where the OS as a whole has a huge amount of software, but you only install what you want/need. In other words, growth of included software in Linux distributions like Ubuntu is not comparable to the same in Windows.

Of course it's also possible that Ubuntu gets more and more software installed by default, but that doesn't make the OS itself bloated, just it's default installation. You can propably remove the unwanted packages from Ubuntu; just try to strip Windows of its bloat.

Re:MS Windows != Every OS

[drsmithy]

...WHA? Yes, there is a concept of a kernel module.

Yes, but kernel modules work by loading into the same address space as the running kernel. Modular drivers (as in NT and OS X) work by passing messages via well defined interfaces.

This is why even a trivial change to the Linux kernel requires recompiling the whole thing (including all the modules) whereas hardware drivers on NT and OS X frequently continue working without modification even through multiple .1-level changes.

I guess that's why I have to download source code to the nVidia driver, patch my kernel, and recompile to make my video card work. Oh wait -- I don't. The nVidia driver gets compiled separately, and produces a module that is loaded on demand. If I don't want to start a GUI, I don't have to load the nVidia module.

Ironically, you've proved his point. You *do* have to download and recompile your nvidia driver every time you get a new kernel, because of how Linux kernel modules work. You do *not* need to do this with Windows or OS X.

On Windows, if I don't want to start a GUI, I'm SOL. Hell, if I don't want IE, I'm SOL.

This is completely and utterly irrelevant to a discussion about kernel architecture.

No. OS X is a hybrid, at least until they decide to nuke the microkernel parts to get a performance gain. Windows is just a monolithic kernel with enough well-defined interfaces that you can easily ship binary drivers that don't screw up the system. Want to prove it's monolithic? Any kind of driver, if it's poorly written, can crash your whole system -- just like on Linux.

Your "proof" is broken (and a driver can just as easily crash OS X - and they do).

Both NT and OS X are so-called "hybrid" kernels. Both were designed with a microkernel architecture and both have subsequently moved away from its strict requirements by running various things in kernel space rather than user space to get better performance.

Both, also, are far closer to a microkernel than Linux is, far more modular and have well defined and stable ABIs.

Windows is screwed up because Microsoft integrates tons of stuff in userland that have no business being integrated.

I challenge you to list some of these things that have been "integrated" on Windows, but not on other platforms.