Building on VMS concepts is basically the same thing as avoiding UNIX concepts. If you want examples, go ask Dave Cutler for an interview, don’t wait for partisans to jump in with their favorite operating system feature. I refer you to http://www.codinghorror.com/blog/archives/000060.html for relevant snippets. See in particular the quote about Dave Cutler’s feelings about UNIX (admittedly second-hand, but probably fairly accurate given what we know about Dave Cutler). It rings fairly true.

How you react to the Dave Cutler quote will show your biases — UNIX folks will say, “Is this guy nuts? These PhD’s all are smarter than Dave Cutler. UNIX has got best-of-breed in every area.” Windows folks will say, “Exactly. I don’t want 37 different design philosophies on 21 programs. Windows is superior in overall consistency, even if it can be easily beat in every single area one-by-one.”

While this is a few days past your post, it is the largest load of unadulterated BS I’ve seen in quite a while. Hell, Windows gets beaten on workstation graphics performance of all things, and that is certainly not what UNIX excels at. Almost any SPEC result is the same — Windows is almost never the top in performance.

Firstly, BSD is not monolithic, and FreeBSD 7 outperforms Linux on up to 8-way SMP loads. Linux is a hybrid kernel (like NT). Mach is a microkernel (OS X sucks on a lot of server loads due to the amount of message passing required).

Secondly, while Windows is many things, “high performance” is not one of them when compared to operating systems optimized for server loads (SQL server aside, which performs well, but is common due to the ease of integration with Windows programs rather than performance characteristics compared to DB2/Oracle/Postgre).

NOBODY uses Vista in a server room. Nor XP, for that matter. Win2k3 is where we’re staying until 2008 gets at least one service pack. That being said, the claim that the NT kernel somehow handles multithreading and multiple processors better than UNIX or Linux is, frankly, laughable. SGI and Cray were running 32/64 processor boxes a decade ago. Take a look at <a href=”http://www.top500.org/stats/list/30/os”?the Top500. Six of the systems are Windows. Six. Two are OS X, which is famously slow for things like that. AIX (IBM’s UNIX, probably on POWER systems, maybe mainframe) has four times as many. Two are Solaris, and you can expect that number to rise now that Sun can cram 128 processors into a 2U box. One is frakking Tru64, and the last major release of that operating system was years ago (IRIX boxes were sitting up there until pretty recently, too, and IRIX has been dead for a while). HP-UX intermittently shows up on the list. All in all, some 450 of the Top500 run… Linux.

As a UNIX sysadmin, your assertion that people install Windows on a large box then install some half-baked POSIX layer and 5 year old non-compliant shell (without Perl or any of the other tools which come -standard- on any UNIX) with the full load of a GUI rather than console-based UNIX/Linux is asinine. Clearly, you’ve never seen a datacenter (or at least one at non-certified Microsoft shop). In the real world, many industries (retail, banking, grocery, scientific) run purely on UNIX for reliability. Banks run HP-UX heavily, grocery is AIX, telco is Solaris. The systems which -absolutely must not break- run UNIX, if you can tell. Hospital gear is QNX and IRIX for the most part on the backend.

XENIX died before NT was a concept, thanks. NT was designed because Microsoft needed a kernel that could support protected memory, multiple processors, etc, and they DID NOT WANT TO PAY AT&T FOR UNIX. It took almost all the concepts from VMS. There weren’t any “flaws” in UNIX they wanted to avoid that I’m aware of.

I will say that NTFS ACLs are much more granular than the general UNIX permissions system. However, all the major UNIXes (Solaris, HP-UX, AIX, Linux, OS X) support far more advanced user roles and monitoring (take a look at SELinux, Trusted HP-UX, or RBAC in Solaris for a comparison) than Windows, and filesystem permissions are hardly the defining role in security (which is really where NT has UNIX beat).

Object-oriented I/O? What does that even mean? As far as it goes, NT does not say “everything is an object.” I can’t cast a file. UNIX says “everything is a file, including devices”, and I can use any tool I want to. It’s far more “object-oriented” than NT with regards to I/O.

UNIX has no I/O limits. Who are the big NASes? NetApp (it’s UNIX on their NASes), EMC (UNIX), Sun (Solaris is clearly UNIX, and ZFS really is the most advanced filesystem out there), etc. Performance pounds Windows NASes and SANs by any benchmark you check on StorageReview.

Message passing through rings is not unique to NT. It’s in every OS since Win9x/DOS (and many before it, such as z/OS, MPE, Nonstop, UNIX, OS/2, VMS, et al). Ring 0 is not a hypervisor.

The perfect “Core OS” as a hypervisor is, by all respects, Linux (VMWare ESX is the virtualization system in the datacenter, and it’s based on RHEL). Even better is something like Logical Domains (Sun), LPARs (IBM), nPARs (HP-UX) which can partition hardware without a clunky hypervisor.

If NT is far more extensible, where are the extensions? How much embedded Windows do you see vs. embedded Linux, QNX, BSD, VxWorks, etc? Virtually none. In the real world, people do not use NT for those tasks. Beyond which, I don’t think you’ve done any development, Windows or otherwise. COM (the big Windows API) is anything but “light.”

Also, Linux/UNIX/BSD do not have a separate video driver model. A driver is a driver, and it gets DMA. X11/Xorg handles graphics and it is unbelievably versatile (as well as extremely modular). To put it this way: X11/Xorg from v6 to v7 was almost a total rewrite. A driver is a driver, and the same “nvidia” driver can drive Xsun (Sun’s proprietary X server), X11 on HP-UX, Xorg 6 on FreeBSD, and Xorg 7 on Linux. It does not take major kernel rewrites. They publish a specification. Driver developers code to the specification. It works, and the specification is stable. Even if the kernel implementation is completely rewritten, the kernel API stays the same. As neat as you may think it is that Windows has two different driver models, driver developers don’t like it as much.

In short, you are utterly wrong. Sure, cleanroom implementations of the NT kernel are flat-out amazing (the 40MB Windows 7 preview for instance). However, even in that case, it is not the ideal solution for some of the cases you’ve named (virtualization, for one). Until I can get a text-mode only version of Windows with a GOOD shell (PowerShell is good for scripting, but if I want an OO scripting language I’m headed for Python or Ruby rather than learning a new one which isn’t at all portable) with modern shell features (in short — port bash, ksh, or zsh rather than throwing away 30 years of perfecting features), it’s relegated to a few things — AD, Exchange, SQL Server, and things which need to integrate with those three.

My 32 processor servers are staying on UNIX or using hardware virtualization, thank you. Come back when you have some professional experience with high-specification servers or kernel internals.

I still have high hopes for Windows 7, and 2008 has some great new features (again, mostly for Domain controllers and Citrix stuff), but it’s not going to displace UNIX/Linux, and arguments for its technical superiority are on shaky ground.

Vendors who make 16 and 32 processor workstations supply a special windows license with the system, and a special install CD that contains a HAL that will work with more sockets.

If I had a BOXX Apexx 8, I would probably run Linux on it.

However, there is plenty of software that one might like to use that can’t run on Linux. Adobe’s After Effects comes to mind as a program that might work well on such a Windows machine, since it is already known to scale well to 8 CPUs.