[Info-vax] OT: Arun Kishan

[John Wallace]

On Jan 16, 4:45 am, John Santos <j... at egh.com> wrote:
> In article <00c9936c$0$23454$c3e8... at news.astraweb.com>,
> jfmezei.spam... at vaxination.ca says...>
>
>
>
> > Neil Rieck wrote:
>
> > > Trying to replace the "Cutler is God" myth with "Windows is OpenVMS
> > > Reimplemented" fact. Don't believe me? Then watch the Rob Short video:
>
> > My old PSION PDA (circa early 1990s) had many features similar to VMS.
> > It had the equivalent to mailboxes for interprocess communications, it
> > had a DECNET like stack where you could remotely start tasks a d
> > exchange data, and you could , on the PSION, edit files residing on
> > another machine.  Priorities were similar as well (any priorty above 16
> > got "real time" like stuff.
>
> > And there were rumours that a number of PSION engineers had come from
> > DEC (Reading).
>
> > Does that make the PDA a "VMS re-implemented ?".
>
> > There are certain core concepts in an OS which are fairly common.
>
> > Windows doesn't have commonality iwt the VMS file system, it doesn'T
> > have RMS, it doesn't have clustering, its file specifications are quite
> > different, its device names are quite different.
>
> Kernel internals (modularity, device drivers, scheduler, memory
> management, etc.) are supposed to be very similar to VMS.  None of
> this is directly visible to users, but is critical for system
> reliability.  
>
> > Let me ask you this: Data General was made up of a lot of ex DEC folks.
> > Does this mean that AOS-VS was a re-implementation of VMS ? It was quite
> > different to the user and system manager, even if some core concepts in
> > the kernel might have been similar.
>
> No!  This would be totally revisionist history.
>
> The DG people left DEC during the early PDP-11 years (1969-1970 or
> there abouts.)  I've heard, but don't know for sure, that the Nova
> design was a 16-bit design very similar to the 12 and 18-bit DEC
> machines, but DEC rejected it in favor of the (at the time) new and
> very different Gordon Bell PDP-11 architecture.
>
> DG started work on their 32-bit system about the same time as, and
> in direct competition to, DEC starting work on the VAX and VMS.
> I am almost totally unfamiliar with AOS, but it can't be a re-
> implementation of VMS since it was written at the same time and
> independently.
>
> The DEC West people who left and went to Microsoft had years of
> VMS experience *before* leaving.
>
>
>
> > It is an insult to say that Windows is VMS re-implememted.
> > Deffectors to Microsoft should not be glorified. People should limit
> > themselves to stating that many ex-Digits went on to work at Microsoft.
>
> --
> John Santos
> Evans Griffiths & Hart, Inc.

"Critical for system reliability" was important in the original NT
design (the one that outsiders saw as NT3.x) but afaik was quickly
(and quietly) sacrificed.

In the original approach, lots of kernel subsystems had their own
separate kernel address spaces and communicated with each other by
"safe" methods such as message passing/procedure calls. (Gross
oversimplification, lots more detail in Custer's "Inside Windows NT"
book).

Obviously because the address spaces are isolated, separate, there is
less chance of unintended data corruption between subsystems, and less
chance of unintended code execution (only explicitly-exported APIs are
accessible). That does indeed enhance system reliability.

But for each call to another subsystem, there is a performance
overhead because of parameter checking and (where necessary) parameter
copying.

Back in the early days of NT, this meant that a given PC doing a
particular set of tasks would likely do them faster on a Win16 OS
(Windows 95/98) than on a Win32 OS (NT3.x/NT4). Gates didn't like
this. Even though the Win32 OS was actually in many cases more
productive (more stable), the "benchmarks" in trade rags showed NT
appearing unacceptably slower than Win95/98.

So to get some of this speed back, Billco dropped the "everything in a
separate address space" model, and more and more kernel stuff that
could/should have been protected was dumped into big shared lumps of
kernel mode stuff. Which got rid of the parameter passing overhead but
of course brought back all the possibilities for unintended data
corruption and unintended data access.

More recently, Vista had a whole load of redesign work done to bring
back some of the isolation which had been abandoned so long ago. This
time the goal wasn't system reliability, the goal was Hollywood-
mandated content protection. Vista (especially Vista64) was supposed
to provide secure tamper-proof end-to-end copy protection of DRM-
protected commercial content such as Blu-Ray. But obviously content
protection is trivially bypassed if any arbitrary kernel mode code can
see other arbitrary bits of data passing through the kernel; what
good's your DRM then? So the "protected media path" was introduced,
along with a whole load of other features that allow media rights
owners to determine what your computer can do with their content.

"System reliability" isn't the driving force behind these changes
though; being a Hollywood-compatible general purpose media-centric OS
is.