Re: ::scr Cesium

[David Cantrell <david@xxxxxxxxxxxxxxx>]

On Tue, Oct 30, 2001 at 11:43:28AM +0000, simon wistow wrote:
> Cesium ...
> 
> It may be a spoof

Yep :-)

>                   but some of the ideas are very nice (not sure about the
> Program layer though, would have to see that in action)
> 
> In a Nut Shell [tm] things I want from an OS (i.e Kernel, API and GUI combo - 
> I'm ignoring certain hardware/laws of physics limitations) :
> 
> o Network transparent 
> For *everything*. I should be able play a game on a monitor and the mouse is
> connected to another machine, the keyboard on yet another machine and the
> processing done by 1 or more machines somewhere. Automagically

Plan 9.

> o Devices should be hot swappable
> I should be able to rip out the graphics card if I want to, put out a new one
> and the OS should keep on running an autoadjust to new settings if necessary

Novell's Unixware (and, I guess, SCO's versions of the same) allowed hot-
swapping of pretty much anything.  I think Vax/VMS does too.  Including
CPUs and memory, given funky enough hardware.  This is, of course, common-
place in the mainframe world.  In (for example) a S390, the only part that
isn't hot-swappable is the back-plane.

> o I shouldn't ever need to reboot
> *Ever* not even when upgrading the kernel or addin, removinf device drivers

Riiiight.  Without a multi processor box with some deep DEEP voodoo that
won't be possible.  You'd need to stop a processor, bring up the new kernel
on that one, then stop all the others and bring up the new kernel on
them.  But there will be a point where you are running both kernels.  And
what of in-memory structures which change from one kernel to the next?
Even in the mainframe world you have to down the machine to replace the
most fundamental pieces of software.

> o Easy Sounds obvious but configuration and location of stuff should follow
> the rule of least surprise. Unix got at least one thing right with its
> 'everything's a file'. Ditto Macs with Folders?

No, Unix got it very wrong by trying *and failing* to make everything be
a file.  Terminals do not behave anything like disks do not behave
anything like files.  Plan 9 is said to do a lot better, but I don't
believe that they could have succeeded.  Better would be to provide as
uniform an interface to everything as possible, but to specifically *not*
try to shoe-horn everything into the same little box.  Unix and I would
get on much better if the OS and the apps would forget about the fiction
that "everything's a file". </rant>

> o Integrated 
> Everything should work together ala OpenDoc and OLE/COM/whatever it's called
> these days. Like piping on *nix but more so - when sending data it should
> automagically present what type of data it is and the recieving app should be
> able to automagically switch to deal with it ...

... or at least be able to fail gracefully.  Or maybe the data should
gracefully degrade.  You're talking throwing objects around with gay
abandon.  I take it that memory bandwidth* isn't an issue then :-)

I'm not sure I like this idea anyway.  It seems to imply that applications
have to work by calling methods on the data objects.  While this is good
for (eg) spreadsheet data, it is not a cure-all.  Sometimes you absolutely
need to get down and dirty.  There will always be a need for that, as
without that ability new application functionality will be slow to develop
as it will require that the data sources are updated before you can deploy
it.  There's also a danger of monopoly control of the data API.  Read "He
Who Controls The Bootloader" and imagine the same applied to Data objects.

* - heh, at least the amount of memory isn't an issue any more :-)

-- 
David Cantrell | david@xxxxxxxxxxxxxxx | http://www.cantrell.org.uk/david

 Pressure was growing last night for the global "war on terror" to be
 broadened to take in a wide range of other 'rogue emotions' including
 horror, shock and a general feeling of bewilderment about the state of
 the world.    -- The Brains Trust