Distributed Computing

Distributed Computing

[Rigved Rakshit]

Hi,

I need help with a couple of things. I will ask them one by one.

I want to "catch" a newly started process, transfer it to a remote machine,
execute this process on the remote machine and then return the result to the
original machine. Can anyone guide me with regard to this? Maybe if anyone
could point me to the relevant documentation, then also that would be great.

One possible method that I thought of is something like this:

1. Process A has just been scheduled to execute for the first time on
machine A.
2. I send a suspend signal to Process A, mostly a SIGSTOP.
3. Then I copy Process A's PCB to machine B.
4. On machine B, I send the continue signal to Process A, mostly a SIGCONT.
5. Whenever Process A needs user input or needs to output something, I cause
a trap, mostly with a SIGTRAP (or I could use SIGSTOP here again). I
transfer this I/O request to machine A and let the user interact. I send the
interaction back to Process A on machine B. I resume Process A on machine B.
6. I continue these steps till Process A has completed execution.

I do not know how to do any of these (but I do know how to transfer files
between two computer using rsync and how to send messages using TCP/UDP). If
someone could point me in the right direction, it would be of great help to
me.

Thanks in advance.

Regards,
Rigved Rakshit
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20110417/57fb85fb/attachment.html 

Distributed Computing

[Mulyadi Santosa]

On Sun, Apr 17, 2011 at 02:04, Rigved Rakshit <r.phate@gmail.com> wrote:
> 1. Process A has just been scheduled to execute for the first time on
> machine A.
> 2. I send a suspend signal to Process A, mostly a SIGSTOP.
> 3. Then I copy Process A's PCB to machine B.
> 4. On machine B, I send the continue signal to Process A, mostly a SIGCONT.
> 5. Whenever Process A needs user input or needs to output something, I cause
> a trap, mostly with a SIGTRAP (or I could use SIGSTOP here again). I
> transfer this I/O request to machine A and let the user interact. I send the
> interaction back to Process A on machine B. I resume Process A on machine B.
> 6. I continue these steps till Process A has completed execution.

Sounds like just what MOSIX, OpenSSI or Kerrighed already do....

-- 
regards,

Mulyadi Santosa
Freelance Linux trainer and consultant

blog: the-hydra.blogspot.com
training: mulyaditraining.blogspot.com

Distributed Computing

[Rigved Rakshit]

>
> > 1. Process A has just been scheduled to execute for the first time on
> > machine A.
> > 2. I send a suspend signal to Process A, mostly a SIGSTOP.
> > 3. Then I copy Process A's PCB to machine B.
> > 4. On machine B, I send the continue signal to Process A, mostly a
> SIGCONT.
> > 5. Whenever Process A needs user input or needs to output something, I
> cause
> > a trap, mostly with a SIGTRAP (or I could use SIGSTOP here again). I
> > transfer this I/O request to machine A and let the user interact. I send
> the
> > interaction back to Process A on machine B. I resume Process A on machine
> B.
> > 6. I continue these steps till Process A has completed execution.
>
> Sounds like just what MOSIX, OpenSSI or Kerrighed already do....
>

Thanks for the information. It was really helpful.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.kernelnewbies.org/pipermail/kernelnewbies/attachments/20110419/66583121/attachment.html