From: gah...@delphi.eng.temple.edu (roger gahman)
Subject: Multi-Processor Development for Linux (PCI bus)
Date: 1995/05/01
Message-ID: <3o2v0b$c37@cronkite.ocis.temple.edu>#1/1
X-Deja-AN: 101751987
organization: Temple University, Academic Computer Services
keywords: Linux, PCI, Multi-Processor, Passive Backplane
newsgroups: comp.os.linux.development.system
summary: Is there any interest in developing a multi-processor version of Linux

I am researching the possibility of developing a multiprocessor version of
Linux using a passive backplane PCI board, with multiple CPU cards.  The main
questions I have on this are the following:

	1.  Is there any interest in such a beast in the Linux community.
	2.  Has anyone looked into this before?  Is there any current work
	    being done in this area?
	3.  For those familiar with the Linux kernel, How difficult do you
	    think this would be?
	
I am an Electrical Engineering student at Temple University, and I am 
considering the possibility of a project along these lines for my Ph.D.
dissertation.  Any suggestions regarding this topic would be greatly 
appreciated.


Roger Gahman
Temple University

From: Ronald Wahl <r...@informatik.tu-chemnitz.de>
Subject: Re: Multi-Processor Development for Linux (PCI bus)
Date: 1995/05/02
Message-ID: <Pine.LNX.3.91.950502015441.174A-100000@goliath.csn.tu-chemnitz.de>#1/1
X-Deja-AN: 101882208
references: <3o35ms$h86@cronkite.ocis.temple.edu>
x-sender: r...@goliath.csn.tu-chemnitz.de
content-type: TEXT/PLAIN; charset=US-ASCII
organization: University of Technology Chemnitz, FRG
mime-version: 1.0
newsgroups: comp.os.linux.development.system

On 1 May 1995, roger gahman wrote:

> I am researching the possibility of developing a multiprocessor version of
> Linux using a passive backplane PCI board, with multiple CPU cards.  The main
> questions I have on this are the following:
> 
> 	1.  Is there any interest in such a beast in the Linux community.
> 	2.  Has anyone looked into this before?  Is there any current work
> 	    being done in this area?

We are working on it since last year. But *at the moment* we don't offer 
information.

An official project called LUMP has just started. Maintainer is Alan Cox 
(iia...@iiit.swan.ac.uk).

> 	3.  For those familiar with the Linux kernel, How difficult do you
> 	    think this would be?

It's very difficult.

 - Ronald

+-------------------------------------------------------------------+
| Ronald Wahl                        r...@informatik.tu-chemnitz.de  |
| >> PGP key available by finger <<  http://www.tu-chemnitz.de/~row |
+-------------------------------------------------------------------+
| PGP fingerprint: 9D 4A 66 7C A9 9F 6A 5F  90 45 D5 0C DF E5 CB 71 |
+-------------------------------------------------------------------+

From: jmo...@mail.scene.nashville.net (Joel Moses)
Subject: Re: Multi-Processor Development for Linux (PCI bus)
Date: 1995/05/02
Message-ID: <3o5ulr$kg3@adam.telalink.net>#1/1
X-Deja-AN: 101882263
references: <3o2lpt$49j@cronkite.ocis.temple.edu>
content-type: text/plain; charset=ISO-8859-1
organization: Nashville Scene/CityPress Publishing, Inc.
mime-version: 1.0
newsgroups: comp.os.linux.development.system

roger gahman (gah...@olympus.eng.temple.edu) wrote:
: I am researching the possibility of developing a multiprocessor version of
: Linux using a passive backplane PCI board, with multiple CPU cards.  The main
: questions I have on this are the following:

That would be a beast, no doubt, but considering the PCI support which has
been added to Linux as of late, might not be a bad idea. I'd love to see
a benchmark of one of those machines against a 4-processor HP. Think I know
which would win, though.

: 	1.  Is there any interest in such a beast in the Linux community.

I'd like to see it.

: 	2.  Has anyone looked into this before?  Is there any current work
: 	    being done in this area?

I know of an effort by a friend to add dual-chip motherboard support for 
Linux. The last I heard, he was running load-balanced P90s quite effectively
and _very_ fast.

: 	3.  For those familiar with the Linux kernel, How difficult do you
: 	    think this would be?

I'll pass on that one.

Joel
j...@moses.com

From: alb...@krakatoa.ccs.neu.edu (Albert Cahalan)
Subject: Re: Multi-Processor Development for Linux (PCI bus)
Date: 1995/05/03
Message-ID: <ALBERT.95May2230541@krakatoa.ccs.neu.edu>#1/1
X-Deja-AN: 101882286
references: <3o2lpt$49j@cronkite.ocis.temple.edu> <3o5ulr$kg3@adam.telalink.net>
organization: Northeastern University, College of Computer Science
newsgroups: comp.os.linux.development.system

>>>>> "J" == Joel Moses <jmo...@mail.scene.nashville.net> writes:

J> I know of an effort by a friend to add dual-chip motherboard support for
J> Linux. The last I heard, he was running load-balanced P90s quite
J> effectively and _very_ fast.

How many crashes/hour under heavy load?  0.2 would be OK for some.
How is the filesystem protected?  NFS would help I think.

J> : 3.  For those familiar with the Linux kernel, How difficult do you :
J> think this would be?

I think the PhD would be well earned.  It would look very good on a resume.
--

Albert Cahalan
alb...@ccs.neu.edu

From: r...@mail.NashScene.com (Nashville Scene Administrator)
Subject: Re: Multi-Processor Development for Linux (PCI bus)
Date: 1995/05/05
Message-ID: <3oe1tb$9ub@adam.telalink.net>#1/1
X-Deja-AN: 102120532
references: <3o2lpt$49j@cronkite.ocis.temple.edu> <3o5ulr$kg3@adam.telalink.net> 
<ALBERT.95May2230541@krakatoa.ccs.neu.edu>
content-type: text/plain; charset=ISO-8859-1
organization: Nashville Scene/CityPress Publishing, Inc.
mime-version: 1.0
newsgroups: comp.os.linux.development.system

Albert Cahalan (alb...@krakatoa.ccs.neu.edu) wrote:
: >>>>> "J" == Joel Moses <jmo...@mail.scene.nashville.net> writes:

: J> I know of an effort by a friend to add dual-chip motherboard support for
: J> Linux. The last I heard, he was running load-balanced P90s quite
: J> effectively and _very_ fast.

: How many crashes/hour under heavy load?  0.2 would be OK for some.
: How is the filesystem protected?  NFS would help I think.

From what he told me, it was stable under heavy load, limited mostly
by disk access speed -- he had to use quite a bit of swap space.

I don't know anything else about it, though. I lost contact with him
a while back.

Joel

From: dabn...@thor.cam.ac.uk (D.A.B. Niggemann)
Subject: Re: Multi-Processor Development for Linux (PCI bus)
Date: 1995/05/07
Message-ID: <3oigqv$lv6@lyra.csx.cam.ac.uk>#1/1
X-Deja-AN: 102196992
references: <3o2lpt$49j@cronkite.ocis.temple.edu> <3o5ulr$kg3@adam.telalink.net> 
<ALBERT.95May2230541@krakatoa.ccs.neu.edu> <3oe1tb$9ub@adam.telalink.net>
organization: University of Cambridge, England
newsgroups: comp.os.linux.development.system

Nashville Scene Administrator (r...@mail.NashScene.com) wrote:
: Albert Cahalan (alb...@krakatoa.ccs.neu.edu) wrote:
: : >>>>> "J" == Joel Moses <jmo...@mail.scene.nashville.net> writes:

: : J> I know of an effort by a friend to add dual-chip motherboard support for
: : J> Linux. The last I heard, he was running load-balanced P90s quite
Are you telling me that this guy actually rewrote the Linux kernel to do fully
_symmetric_ multiprocessing (with all the reentrancy and locking issues this 
implies), or was he just allowing user processes to run on another processor, 
with blocking on kernel entry, that is, an asymmetric multiprocessor system 
with all OS requests handed on one processor or ping-ponged between both 
processors but never executing in parallel on both? The latter would indicate 
an impressive hacking skill, the former means someone should give this guy 
a permanent reference amongst the linux.gods.....
: : J> effectively and _very_ fast.
Has this guy posted his modifications 
etc to anyone, anywhere? 
: : How many crashes/hour under heavy load?  0.2 would be OK for some.
: : How is the filesystem protected?  NFS would help I think.

: From what he told me, it was stable under heavy load, limited mostly
: by disk access speed -- he had to use quite a bit of swap space.

: I don't know anything else about it, though. I lost contact with him
: a while back.
If you could find this guy again, you would probably be doing the Linux 
community one of the biggest favours, ever!(Dreaming of 4-processor 
P120's for < $4000, _drool_, sorry got a bit caried away) 
: Joel


--
__________________________________________________________

 |  /  _ \ _ _| __ \          Dirk Niggemann 
 ' /  |   |  |  |   |         Jesus College
 . \  __ <   |  |   |         Cambridge, CB58BL
_|\_\_| \_\___|____/          dabn...@cam.ac.uk
__________________________________________________________

From: h...@mathi.uni-heidelberg.de (Hannes Reinecke)
Subject: Re: Multi-Processor Development for Linux (PCI bus)
Date: 1995/05/09
Message-ID: <HARE.95May9145841@mathi.uni-heidelberg.de>#1/1
X-Deja-AN: 102304835
references: <3o2lpt$49j@cronkite.ocis.temple.edu> <3o5ulr$kg3@adam.telalink.net>
organization: University of Heidelberg, Germany
reply-to: h...@mathi.uni-heidelberg.de
newsgroups: comp.os.linux.development.system

=>"D" == D A B Niggemann <dabn...@thor.cam.ac.uk> writes:

D> Nashville Scene Administrator (r...@mail.NashScene.com) wrote:
D> : Albert Cahalan (alb...@krakatoa.ccs.neu.edu) wrote:
D> : : "J" == Joel Moses <jmo...@mail.scene.nashville.net> writes:

D> : : J> I know of an effort by a friend to add dual-chip
D> : : J> motherboard support for Linux. The last I heard, he was
D> : : J> running load-balanced P90s quite 
D> Are you telling me that this guy actually rewrote the Linux kernel
D> to do fully _symmetric_ multiprocessing (with all the reentrancy
D> and locking issues this implies), or was he just allowing user
D> processes to run on another processor, with blocking on kernel
D> entry, that is, an asymmetric multiprocessor system with all OS
D> requests handed on one processor or ping-ponged between both
D> processors but never executing in parallel on both? The latter
D> would indicate an impressive hacking skill, the former means
D> someone should give this guy a permanent reference amongst the
D> linux.gods.....  
D> : : J> effectively and _very_ fast.  
D> Has this guy posted his modifications etc to anyone, anywhere?  
D> : : How many crashes/hour under heavy load?  0.2 would be OK for
D> : : some.  
D> : From what he told me, it was stable under heavy load, limited
D> : mostly by disk access speed -- he had to use quite a bit of
D> : swap space.
D> : I don't know anything else about it, though. I lost contact
D> : with him a while back.  
D> If you could find this guy again, you would probably be doing the
D> Linux community one of the biggest favours, ever!
And it would stop me from disbelieve. Until somebody is able to
contact this guy and verify that, I'd suggest in forwarding this to
net.legends. 
I _know_ of a group at cs.uni-chemnitz.de who is trying to port linux
to a multiprocessor pentium. They have started last year and the port
is still _very_ alpha. In fact, they are very short of comments, so I
didn't know the current state. 
But they are working on it.

Have fun

Hannes
-------
Hannes Reinecke			     |
<h...@vogon.mathi.uni-heidelberg.de> |  XVII.: WHAT ?
				     | 	
PGP fingerprint available            | 		T.Pratchett: Small Gods
see  'finger' for details	     |		

From: Ronald Wahl <r...@informatik.tu-chemnitz.de>
Subject: Re: Multi-Processor Development for Linux (PCI bus)
Date: 1995/05/10
Message-ID: <Pine.LNX.3.91.950509234546.893A-100000@goliath.csn.tu-chemnitz.de>#1/1
X-Deja-AN: 102304853
references: <3o2lpt$49j@cronkite.ocis.temple.edu> <3o5ulr$kg3@adam.telalink.net>
x-sender: r...@goliath.csn.tu-chemnitz.de
content-type: TEXT/PLAIN; charset=US-ASCII
organization: University of Technology Chemnitz, FRG
mime-version: 1.0
newsgroups: comp.os.linux.development.system

On 9 May 1995, Hannes Reinecke wrote:

> I _know_ of a group at cs.uni-chemnitz.de who is trying to port linux
> to a multiprocessor pentium. They have started last year and the port
> is still _very_ alpha. In fact, they are very short of comments, so I
> didn't know the current state. 
> But they are working on it.

The maintainer haven't the time to read news and so I will do it. But at 
the moment I'm also short of time to comment each article. We have a 
documentation (only partiell for now) to our project. It should be soon 
available as html'ed ps-file under 
http://www.tu-chemnitz.de/~jwern/parlabs.html. But it's only in german.
But on request I can send it via mail.
The current state is that we have implemented kernel supported threads and
the relating protection stuff like mutexes, condition variables and
semaphores. The base kernel version is 1.0.9. But we can't offer a public 
version yet because the hardware stuff is Compaq specific and at the moment 
we haven't the support from this company but we are in negotiation. If we 
are certain that we get the support then we maybe publish it. Maybe 
sombody who knows some people at compaq can help us.

There are also two other projects:

 * LUMP maintained by Alan Cox (iia...@iiit.swan.ac.uk), which has also a 
   mailing list.

 * Project Viper. But the authors of Viper have very low time to work on 
   it now.  

Ciao,
  Ronald.

+-------------------------------------------------------------------+
| Ronald Wahl                        r...@informatik.tu-chemnitz.de  |
| >> PGP key available by finger <<  http://www.tu-chemnitz.de/~row |
+-------------------------------------------------------------------+
| PGP fingerprint: 9D 4A 66 7C A9 9F 6A 5F  90 45 D5 0C DF E5 CB 71 |
+-------------------------------------------------------------------+