Re: Intel details future Larrabee graphics chip

On Mon, 11 Aug 2008 12:17:07 +0100, Martin Brown
<|||newspam|||@nezumi.demon.co.uk> wrote:

John Larkin wrote:
On Fri, 08 Aug 2008 18:12:14 +0100, Martin Brown
<|||newspam|||@nezumi.demon.co.uk> wrote:

John Larkin wrote:
So what do you think OS's will look like 10 years from now, when even
home computers run on chips with 100's of cores? Still one gigantic
VMS/Mach/NT descendent running on one CPU, thrashing and piping

A lot of IO is concentrated by the bridge hardware these days. And
serial ports have had moderate to large FIFOs for about a decade.

XP runs quite happily on my dual core. Vista runs less happily on my new
Toshiba portable and I will never recommend using it to anyone.

XP should be fairly stable by now. Its code base been continuously
debugged and repaired for about 15 years now. Probably 50K bugs have
been fixed.



And those other cores stay idle unless you play a game?

I can see a case for cores allocated to processes with highest demand
for resources, but I do not believe it makes any sense to have one
thread per core with a properly designed secure operating system.

So what do you do with all those cores? Let N-1 of them sit idle, and
multitask/context switch/swap like mad one the one you do use?



Umm, excuse me, what do those words mean, "properly designed secure
operating system" ?

They mean one that uses the security features properly that already
exist in many CPU architectures (and on properly designed hardware that
is not vulnerable to tricks to obtain engineering diagnostic mode or
executing bogus partially decoded non-existent instructions to trap and
claim ring0 status). But alas the hardware isn't perfect either eg.

http://msowww.anu.edu.au/~peterson/pentium_lock.htm

That's what my wife asked me once when I was stupid enough to use the
phrase "too much garlic."


In exactly the same sense as you claim for your magical hardware
architecture a properly designed secure OS would be well secure.

There's nothing magical about lots of cores. Everybody is doing it.

Everybody is doing lots of cores, but very few are advocating the
wasteful and naive usage strategy that you seem to want.


No, the new new thing is hypervisors and virtualization. Programmers
never want to simplify, because that's no fun. They want to abstract,
to pile complexity on top of the complexity they've grown bored of.

A microkernel, written by one programmer, that runs on one CPU, that's
fully documented and never breaks, is indeed naive. It's naive because
the current computing culture would ridicule it as being
insufficiently clever.



I could be persuaded that Mickeysoft leave 'Doze vulnerable to avoid
putting the AV people out of business (that would be anti-competitive).

As James says, don't assume malice when incompetance will do.

I suspect you are right. But the conspiracy theory is more fun!

Things will never change? We'll always use 1980's OS architectures?

Sadly I suspect that might well be the case until some compelling reason
to change comes along. Do you not remember how long the delay was before
there were 32bit consumer grade OS's for the early 386 PCs?

What may well happen is that, once hundred-core CPUs are out in the
wild, some small group of Linix kernal jocks will spin a version that
*can* have file systems, drivers, stacks, and apps assignable to
various CPUs. Then it would just be a configuration thing to assign
one cpu to run just the OS. That would be dynamite for server apps.

Then Microsoft will scramble to catch up, as usual.

A decent OS properly configured doesn't crash all that often (even when
doing software development). My XP box crashes maybe once every few
months, my old Win98 box slightly more often. These days it needs a good
kick to free the disk stiction if I switch it on. Win95 crashed at least
daily. But NT3.51 was a gem, and my OS/2 box only ever crashed about
once a year (perhaps I remember them through slightly rose tinted glasses).

Reducing crashes is actually the easy part. Security is a tougher
problem. Windows is the worst, but AppleOS and Linux have security
problems, too. This should simply not be possible.



My new Vista on a Toshiba portable crashes (or rather its keyboard and
on/off switch locks up) every time I leave leave it on and unattended
for long enough that the power managment engages. Sadly it doesn't have
a reset switch (my bad). I would not use Vista from choice but customers
want me to :(

I do agree with you that consumer OS's have become far too big and
clunky with bloatware and not enough emphasis on security. Where we
differ is on what to do about it.


I vote to simplify. What do you want to do?

John


.