Mail Archives: opendos/2000/12/26/02:19:15
Hi Stephen,
----- Original Message -----
From: "Stephen Kitt" <lists AT sk2 DOT org>
To: <opendos AT delorie DOT com>
Sent: Monday, December 04, 2000 4:46 PM
Subject: Re: BASIC & EMS (was: Optimizing CONFIG.SYS...)
> Multi-processing only refers to the physical architecture of a system, ie
> the fact that it has multiple CPUs (read Hennessy & Patterson for details,
> or is it Hwang...). Multi-tasking refers to the ability to run multiple
> tasks "simultaneously". The difference between MP and uni-processor
systems
> is that MP systems can really run multiple tasks simultaneously, up to the
> number of processors in the system, although performance gains aren't
linear
> due to the overhead of managing multiple processors.
Yes this is ture. In fact you do not even have to have all CPU's in one
location. They can be loacted in different physical loacions. Also many
systems and all 386/486 (that I am aware of) use plugin CPU cards on a
backplane. The first multi CPU MBs I saw were Pentiums.
As for gains not being linear, this is also true. For the PCs, you don't
gain that much after the first four. If you go to eight CPUs you will not
double the troughput. This is noticable on multiple CPUs on a single MB.
> There are super-computers with loads of CPUs all over the place. You're
> probably thinking of the University of Edinburgh's Parallel Computing
> Centre, which has a number of Cray super-computers, including a T3E which
> currently has 348 processors. Or you might be thinking of Transputers,
which
> could be assembled to form super-computers with many processors. Thinking
> Machines built super-computers using them, and EPCC had one for a while.
> Have a look at http://www.top500.org for more information on
> super-computers, and http://www.epcc.ed.ac.uk for more information on
EPCC.
> Currently, the fastest super-computer in the world (IBM's ASCI White) has
> 8192 processors...
The one I was thinking of used 8192 DEC Alphas. I don't have the original
information about it handy. It's on a tape that was used to store Linux
stuff and I cannot access those tapes at this time. I believe it was in
Scottland where they built this super computer.
> Incidentally, Linux supports thousands of processors on some
architectures,
> and has done so for a while - it was ported to a Fujitsu super-computer a
> while back.
I was just speaking of the PC version. I think it was limited to 4 CPUs, but
could have been 8, because someone had shown that 8 was not that much better
than 4. But they may not have using Linux. There was an OS and it may still
exist called VM386 as I recall that could use multiple CPUs. These were used
on a backplane with plugin CPU cards. the company that I talked to about
these also sold the hardware. I may even have some of that documentation
around here somewhere. This was before the Pentiums had even come out. So it
was quite some time ago. As I recall you could run DOS apps on this OS. It
was a multi-user, multi-proccessor OS. I was mainly interested in it to be
used for using terminals instead of PCs connected together to be used by
small businesses. I found some lower cost alternatives for what I needed.
> There isn't much difference between Desqview, Task Manager and Linux when
it
> comes to the essentials of multi-tasking. Every one of them time-slices...
> Linux handles idle tasks much better, thanks to the way Unix software is
> written, but DV and TM have many tricks up their sleeve to cope with idle
> software too. In fact, when an application looks for input from the
keyboard
> it is considered idle by DV and TM, assuming it uses DOS services to
access
> the keyboard and doesn't do it directly. Check the difference between say
> 4DOS waiting at the prompt and LIST waiting for input - LIST is
notoriously
> uncooperative. (I'm skipping some details here, particularly with regard
to
> Linux; the latter has benefitted from lots of development which DV and TM
> haven't, and in particular its scheduler makes a huge difference.)
I am not certain what tricks you are talking about with TM and DV. I did
have a bunch of 3rd party stuff for DV. TM I set things up in the INI file.
I don't even modify that very much.
I certainly can tell the difference between Linux and any other multitaking
OS or multitasker I have ever used. It is much faster and nothing crashed.
The same might be true with OS/2, but unfortunately, every time I got
supposed software for OS/2, it was some WINDOZE 16 bit app. (I used the
supplied WINDOZE kernel that cape with OS/2 Warp 3.0 Plus Bonus Pack.) But
with what few things I did have for it, it did seem a lot better than the
other stuff I have used. NT is not too much better than WINDOZE. It is just
a lot more stable. NT just does not work like Linux does.
> As much as you may dislike it, Windows does the same - DOS tasks have been
> time-sliced since Windows 3.0 (in extended mode), and Win32 tasks are
> time-sliced in Windows 95 and later. NT time-slices everything if I
remember
> correctly...
>
> From some of your other emails I've got the impression (I may be wrong
here)
> that you seem to think V86 mode helps multi-task programs. The
> virtualisation here is only helpful in that it creates isolated PCs; it
> doesn't run them, the task scheduler still has to do the time-slicing.
> Virtualisation on x86 CPUs is nothing like that offered on mainframes such
> as IBM's S/390.
You are wrong about that, you do not have to use V86 to multitask nor do you
have to multitask when using V86. It's advantage is just as you say. each
task is isolated. You can have one v86 machine crash and still have all the
others running. You can put a single task into a single v86 and run several
tasks each on it's own v86. Each v86 could also run multiple tasks, but I
see no reason to do it. You could for example run DV in a TM window. I
never tried it, but seem to recall that it can be done. You could also run
WINDOZE 3.3 as a task under TM. I never tried it.
> Before you argue further about memory managers, standards, and
> multi-tasking, I strongly suggest you go and read the technical
> specifications for Intel CPUs and memory management standards. Oh, and try
> writing some code to use the stuff - that often helps clear ideas...
The memory manager is really hardware. The rest is software to utilize it.
There are probably as many ways to use memory management as there are
programmers. Each programmer could do whatever they wish.
Pat
_________________________________________________________
Do You Yahoo!?
Get your free @yahoo.com address at http://mail.yahoo.com
- Raw text -