openMosix

i'm working on my final project, and i successfully installed an =
openmosix cluster. you probably won't like to hear this, but now have to =
do the same with micro$oft windows. has anybody an idea how to obtain =
process migration on a windows machine?

thx

jeroen

It's doable. You first have to obtain the Windows source code. Then,=20
obtain the permit to modify it. Then, understand the source code and=20
write source migration patches for it. Then, obtain the compilation=20
environment from Microsoft and pay for it. Then, obtain permission from=20=

Microsoft to distribute, sell, or otherwise make available to third=20
parties your changes without breaking the customer's support from=20
Microsoft.

I guess the shortest task is the third one.

Moshe


On Tuesday, May 13, 2003, at 07:02 US/Pacific, jeroen degadt wrote:

> i'm working on my final project, and i successfully installed an=20
> openmosix cluster. you probably won't like to hear this, but now=A0have=20=

> to do the same with micro$oft windows. has anybody an idea how to=20
> obtain process migration on a windows machine?
> =A0
> thx
> =A0
> jeroen

Windows 2000 Advanced Server can do it i guess
(2000$ :)
Marc-o


    i'm working on my final project, and i successfully installed an
openmosix cluster. you probably won't like to hear this, but now have to do
the same with micro$oft windows. has anybody an idea how to obtain process
migration on a windows machine?

    thx

    jeroen

Jeron-
 As far as I know there is _no_ SSI software for windows and very little clustering software in general. The other bad thing I get have to tell you is if you do find anything it will likely cost a bundle. The one hope I can give you is there was some kinda of hardware level clustering hack for the X-box. 

Evan 

On Tue, 13 May 2003 16:02:25 +0200
"jeroen degadt" <gva02@...> wrote:

> i'm working on my final project, and i successfully installed an openmosix cluster. you probably won't like to hear this, but now have to do the same with micro$oft windows. has anybody an idea how to obtain process migration on a windows machine?
> 
> thx
> 
> jeroen
>

On Tue, 13 May 2003 07:31:19 -0700
Moshe Bar <moshe@...> wrote:

> It's doable. You first have to obtain the Windows source code. Then,=20
> ... snip ...

LOL!!! This is one of the best quotes from Moshe I've read so far :)

Kris, maybe we should add a "Moshe Bar quotes from the mailing lists" secti=
on
in the oM-HOWTO ;)

Or maybe Bruce can put up a web page for that ;)
 =A0
Seriously though, Jeroen, I'm happy you've setup an openMosix cluster. As
everyone here said, clustering under Windows is a very weak side of the OS =
af
far as we know. That's why we love openMosix. You can get it for free, play
with it and even use it in a production system. And you have the _full_ sou=
rce
code which you can modify/fix and it's all free as in beer + speech.

You cannot expect a serious reply about "Windows clustering" from this list.
Maybe just ask Microsoft Support or google a bit about the topic.

If anyone has any good suggestion, even about a commercial product, I'd be =
real
glad to hear about it. At least we can understand what does the "competitio=
n"
look like in the SSI clustering market today.

Ciao, Mirko.

Mirko Caserta wrote:

> You cannot expect a serious reply about "Windows clustering" from this list.
> Maybe just ask Microsoft Support or google a bit about the topic.

  Oh go on, I'll try and give a serious reply!

  The nearest thing to openMosix on Windows would probably be the PVM 
port to Windows. It's not process migration, but it does allow spare 
cycles sitting around on Windows boxes to be used up. You need PVM-aware 
code and/or libraries.

  No chance of me running it here though - our Windows boxes fall over 
enough without it!

Baz

> Jeron-
>  As far as I know there is _no_ SSI software for windows and very little clus
> tering software in general. The other bad thing I get have to tell you is if 
> you do find anything it will likely cost a bundle. The one hope I can give yo
> u is there was some kinda of hardware level clustering hack for the X-box. 

A solution maybe to get openmosix working on user-mode-linux running on win32.

:)

ach

On Tue, 13 May 2003 20:17:21 +0100
Andy Hayward <ach@...> wrote:

> > Jeron-
> >  As far as I know there is _no_ SSI software for windows and very little clus
> > tering software in general. The other bad thing I get have to tell you is if 
> > you do find anything it will likely cost a bundle. The one hope I can give yo
> > u is there was some kinda of hardware level clustering hack for the X-box. 
> 
> A solution maybe to get openmosix working on user-mode-linux running on win32.
> 
> :)
> 
> ach

openMosix runs great under VMWare (if you can stand the price). maybe if bochs or some other open-source x86 emulator runs on windows...?

On Tue, 13 May 2003 16:59:14 -0400
Peter Willis <psyphreak@...> wrote:
 Don't think Vmware worked as a node. I thought it just migrated well.

Evan

> 
> openMosix runs great under VMWare (if you can stand the price). maybe if bochs or some other open-source x86 emulator runs on windows...?
> 
> 
> ---------

On Tue, 13 May 2003 17:29:30 -0400
"Miguel A. Mota Jr." <Chaos325@...> wrote:

>     When I run mosmon, no nodes appear.  The openmosix status query returns
> that openMosix is not started.  I start it and I get the same message from
> above.  Still no nodes appear.  openMosixview also fails to detect the
> nodes.  Is this a problem with Red Hat 9 or am I doing something wrong here?

You have probably installed the 0.3.1-1 rpm package which I made and soon hid a
few days later just because I discovered this problem.

The solution is to either downgrade to openmosix-tools-0.3-1.i386.rpm:

http://prdownloads.sourceforge.net/openmosix/openmosix-tools-0.3-1.i386.rpm?download

or you should need to have a copy of the Linux Kernel 2.4.20 with the
openMosix-2 patch applied into the /usr/src/linux-openmosix directory. Do a
make menuconf + dep into the main Kernel dir, then get the
openmosix-tools.src.rpm at

http://prdownloads.sourceforge.net/openmosix/openmosix-tools-0.3-1.src.rpm?download

Next you have to:

# rpmbuild --rebuild openmosix-tools-0.3-1.src.rpm

While rebuilding, the tools will look for the matching kernel headers in
/usr/src/linux-openmosix and, assuming your "make menuconfig" is correct, you
should end up with an /usr/src/redhat/i386/openmosix-tools-0.3-1.i386.rpm

If rpmbuild doesn't work, simply download the openmosix-tools tarball:

http://prdownloads.sourceforge.net/openmosix/openmosix-tools-0.3.tgz?download

unpack it somewhere and just type "# make" as root. If you have a properly
configured Kernel src dir under /usr/src/linux-openmosix it should work. If you
use the tarball, see what name the ncurses based mon.c program has in the "mon"
directory, that's because the RH install calls it mmon, while the default for
the tarball is to call it mosmon (but I may be wrong here and maybe confusing
Debian with RH).

Hope this helps. At least it has worked for me when I had the same problem.

I might have said some very wrong thing here so anybody (Moshe?) please correct
me if I'm wrong.

What I can say is that the latest stable releases are those you can see in the
official oM's website at: http://openMosix.sourceforge.net/

Mirko

PS: last minute update... I've just received an email from Cristiano saying he
has committed major changes to the userspace-tools module in the CVS repository
and the well-known procedure:

$ ./configure
$ make
# make install

should work exactly like the 0.3 one. Also try and see what 
"./configure --help" suggests ;)

Eventually, the configure script will disappear from CVS, as well as the debian
dir but all this needs to be discussed with Moshe and the userspace-tools
developers and maintainers.

i am currently not on a Linux box as I am travelling, but I assume 
Berkeley DB uses mmap for the DB files. If it tries to open with mmap 
shared it will fail under oMFS.

Moshe


On Tuesday, May 13, 2003, at 10:01 US/Pacific, Gordon Tisher wrote:

> Hi there,
>
> I'm having problems with Berkeley DB files and MFS -- any program that 
> uses db files doesn't work.  The current example is the vacation 
> program.
>
> I'm running Slackware 8.1, with a custom kernel built from plain linux 
> 2.4.19 and openMosix 2.4.19-7.
>
> Any info on this?
>
> Thanks,
>
> Gordon
>
> -- 
> Gordon Tisher
> Senior Developer
> Gavagai Technology
>
>
>
> -------------------------------------------------------
> Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara
> The only event dedicated to issues related to Linux enterprise 
> solutions
> http://www.enterpriselinuxforum.com
>
> _______________________________________________
> openMosix-general mailing list
> openMosix-general@...
> https://lists.sourceforge.net/lists/listinfo/openmosix-general
>

if all you want to do is migrate matlab processes, install the migshm=20
patch for openMosix.

M


On Tuesday, May 13, 2003, at 09:08 US/Pacific, jeroen degadt wrote:

> i'm trying to migrate matlab 6.5 processes, but i've read openmosix=20
> can't deal with this.
> this is the main reason why i'm looking for an alternative on windows.
>
> ----- Original Message -----
> From: Moshe Bar
> To: jeroen degadt
> Cc: openmosix-general@...
> Sent: Tuesday, May 13, 2003 4:31 PM
> Subject: Re: [openMosix-general] (no subject)
>
> It's doable. You first have to obtain the Windows source code. Then,=20=

> obtain the permit to modify it. Then, understand the source code and=20=

> write source migration patches for it. Then, obtain the compilation=20
> environment from Microsoft and pay for it. Then, obtain permission=20
> from Microsoft to distribute, sell, or otherwise make available to=20
> third parties your changes without breaking the customer's support=20
> from Microsoft.
>
> I guess the shortest task is the third one.
>
> Moshe
>
>
> On Tuesday, May 13, 2003, at 07:02 US/Pacific, jeroen degadt wrote:
>
> i'm working on my final project, and i successfully installed an=20
> openmosix cluster. you probably won't like to hear this, but now=A0have=20=

> to do the same with micro$oft windows. has anybody an idea how to=20
> obtain process migration on a windows machine?
> =A0
> thx
> =A0
> jeroen
>

On Mon, 12 May 2003 19:03:38 -0700
Moshe Bar <moshe@...> wrote:

> Yes, I had thought of that, too. We'd need to have the init scripts for 
> openMosix set the migration capabilities. Mirko, Ettore, Matt?

Yes, I have a Gentoo on my notebook and I can understand what tantive says.
I'll need to have a closer look at how clustermask works and see if we can
automate the thing by using some defaults in the often misunderstood
/etc/openmosix/openmosix.config file (which is not the openmosix.map file!) ;)

In a perfect world, we could use gomd's features to ask other nodes about their
CPU Instruction Set flags and make the clustermask changes accordingly but, for
the moment, the gomd project is still in a very alpha release and I guess we'll
need to handle it in the aforementioned config file.

Any suggestions? Ettore? Matt?

Ciao, Mirko.

Hey all-
  I was about to run into this problem my self. I was thinkabout something along the lines of exporting the /proc/cpuinfo for comparison and then masking the the nodes with the same feature sets together. On a small cluster or very broad range  of hardware I think it will still make more sense to compile code the spec of the weakest machine.

Evan 


On Tue, 13 May 2003 10:24:15 +0200
Mirko Caserta <mirko@...> wrote:

> On Mon, 12 May 2003 19:03:38 -0700
> Moshe Bar <moshe@...> wrote:
> 
> > Yes, I had thought of that, too. We'd need to have the init scripts for 
> > openMosix set the migration capabilities. Mirko, Ettore, Matt?
> 
> Yes, I have a Gentoo on my notebook and I can understand what tantive says.
> I'll need to have a closer look at how clustermask works and see if we can
> automate the thing by using some defaults in the often misunderstood
> /etc/openmosix/openmosix.config file (which is not the openmosix.map file!) ;)
> 
> In a perfect world, we could use gomd's features to ask other nodes about their
> CPU Instruction Set flags and make the clustermask changes accordingly but, for
> the moment, the gomd project is still in a very alpha release and I guess we'll
> need to handle it in the aforementioned config file.
> 
> Any suggestions? Ettore? Matt?
> 
> Ciao, Mirko.
> 
> 
> 
> -------------------------------------------------------
> Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara
> The only event dedicated to issues related to Linux enterprise solutions
> http://www.enterpriselinuxforum.com
> 
> _______________________________________________
> openMosix-general mailing list
> openMosix-general@...
> https://lists.sourceforge.net/lists/listinfo/openmosix-general
>

Hi all!


barnowl@... wrote:
>Hey all-
>I was about to run into this problem my self. I was thinkabout something along the lines of exporting the /proc/cpuinfo
 >or comparison and then masking the the nodes with the same feature 
sets together.
>On a small cluster or very broad  range of hardware I think it will still make more sense to compile 
code the spec of the weakest machine.

yes, that sounds good! But don't forget that the slow machines (the ones 
with less features) should be able to use the fast ones for migration.
All that has to be avoided is that code optimized for a specific flag 
(e.g. sse2) is run on a machine not having support for that flag.

I don't think it makes sense to compile code optimized to the specs of 
the weakest machine. (But that's another discussion)


Regards
Michael

I think Louis has been working towards such a feature. Louis?

Moshe


On Monday, May 12, 2003, at 16:03 US/Pacific, Michael Imhof wrote:

> Hi all!
>
> It's me again as some other feature came to my mind.
>
>
> As we have autodicovery available for nodes (which works great), what 
> about a similar tool that handles complex network topolgy?
>
> That came to my mind as i upgraded today some machines from 100mbit to 
> gigabit network and the machines with the slow network couldn't see 
> the ones with gigabit and vice versa.
>
> Having such a daemon would make setup and handling of 
> openmosix-clusters even better!
>
>
> Comments are welcome.
>
>
>
> Regards
> Michael
>
>
>
> -------------------------------------------------------
> Enterprise Linux Forum Conference & Expo, June 4-6, 2003, Santa Clara
> The only event dedicated to issues related to Linux enterprise 
> solutions
> http://www.enterpriselinuxforum.com
>
> _______________________________________________
> openMosix-general mailing list
> openMosix-general@...
> https://lists.sourceforge.net/lists/listinfo/openmosix-general
>