[lkml]   [1997]   [Nov]   [4]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: Linux + Win95 simultaneously
On Mon, 3 Nov 1997, Byron Davies wrote:

> Second, here's why I'm interested. There are a lot of people (e.g., in the
> engineering community) who need Linux to do serious computing and Windows
> to share the results.

Then get applixware -- still cheaper than Windows plus modified
multiprocessor box.

> These are people who are pro-Linux (for its
> capabilities) but not necessarily anti-Microsoft. Dual booting is a major
> pain and having two separate computers is expensive. Emulation is a great
> idea, but it's not easy either. (Through the beauty of freeware, however,
> it's likely that a substantial amount of Wine could be recycled into a dual
> processor, dual OS system, whether implemented via PCI or shared memory.)

Shared PCI _and_ shared memory -- unless you have separate buses and
memory for processors that effectively will make two motherbords. There is
no way to teach Windows share PCI bus, and even less likely share devices
like VGA and hard drive. The closest thing I have seen was Marathon
"cluster" where two Windows boxes with motherboards had all their
peripherials handled by two another Windows boxes, and those second boxes
synchronized first ones, so two processors executed the same code
simultaneously for redundancy (yes, I agree, there are few things with
more idiotic design). One may try to use the same kind of design to make
Windows box run through Linux box using it as the complete I/O handler
(similar to how Citrix handles graphics and input devices through X), but
the effort needed and the resulting design that will contain two complete
or almost complete computers don't make any sense for me.

> My customers would be ecstatic if they could, as Robert G. Brown put it,
> hot-key between Linux and Windows.

Place two motherboards into one box, connect them by ethernet, make
Windows one diskless and add mechanical relay to switch monitor, mouse and
keyboard by signals drom parallel ports on Linux motherboard. The same
result, saves power supply and disks, but way less headache.

> Performance wouldn't be a major issue.
> From what I've read, the dual processor Pentium board provides only
> marginally better performance anyway.

It depends on SMP implementation in OS and the kind of tasks, box is used
for. What you are suggesting is not SMP in the slightest degree -- it's a
box where two completely separately systems for some insane reason use the
same bus and memory. I don't see any reason for sharing resources that are
not common in their nature (this is my main complaint about multithreading
-- programmer is forced to share and maintain locks on resources that in
multitasked programs are just completely unrelated), so I think, second
motherboard will be much better idea. There are some commercially
available boards that are plain PCI/ISA backplanes without anything on
them, and motherboards that are just cards for such backplanes, so
building a box with multiple motherboards in a "baby-tower" case shuldn't
be a big problem.

> What I would hope from this
> configuration is super convenience at a small incremental cost.

The question still is -- why?

> How would RAM be shared? Run Windows in low memory, relocate Linux to a
> higher region.

It's absolutely unreasonable -- bus sharing between systems thar know
nothing about each other and have drivers that can't cooperate at all will
be insane.

> I'm brand new to Linux myself, having been a Mac user for 5
> years and a Lisp machine user for 12 years before that, so I have no idea
> if this is possible.
> Why not a Windows app precompiler, as Robert G. Brown suggests?

Because the whole point of running Windows is (if such thing exists) to
run applications that are available only in binary, most likely only for
Intel, and no source is going to be available for porting or recompiling.

> This
> sounds similar to interpreting Java byte-codes. This approach is
> equivalent in complexity to emulation, though it has greater opportunities
> to apply techniques of compiler optimization.

Interpreting is used in WABI for non-Intel platforms. Still WABI includes
parts of Windows, so the point is moot.

> Once Wine works, then
> Robert's technique can be used to speed things up.
> Yes, there is a lot of work to making such a system work, but if you know
> of a technical reason that makes it impossible, please tell me.

It's much easier to make dual-motherboard box. Still, sice buying
anything from Microsoft or distributing anything in Microsoft proprietary
formats I consider to be unethical, I am not going to use such thing,
although multiple-motherboard box can be useful in clusters.


 \ /
  Last update: 2005-03-22 13:40    [from the cache]
©2003-2011 Jasper Spaans