Hi,
i am working on product which uses the MPI on Linux.
I have some working knowledge of MPI.
If any one is interseted we can continue the discussion on this forum or they
can contact me personally.
Thank you,
Srinivas.A.V.
Arun Sharma <[EMAIL PROTECTED]> on 24/03/2000 05:48:55 AM
Please respond to [EMAIL PROTECTED]
To: [EMAIL PROTECTED]
cc: (bcc: A V Srinivas/Systems/NSEIL)
Subject: Re: [LIH] one code- n machines
On Thu, Mar 23, 2000 at 08:26:18AM -0800, Nick Hill wrote:
Don't know much about Beowulf and MPI. But to answer a specific
question below:
> I know the memory sharing is not possible when the systems are
> seperated, but is there any way that multi-threaded progs, will run
> UNMODIFIED on clusters, making adequate use of all the systems. And what
> level of Ethernet is reuquired? 10mbps is fine, or do we have to look
> for higher speeds. And is network boot along with NFS ESSENTIAL for a
> cluster? or can i just take individual systems, and put them all into
> some cluster mode after booting? I know they're a lot of questions.
> I've been saving them for a loooooong time, since i dont have anybody
> to ask here. Sorry for that...
Yes, this kind of work has been done in research - emulating SMP on
a cluster. The technique is called Distributed Shared Memory (DSM).
You play tricks with page table entries and fault handlers communicate
over the network.
http://www.cs.umd.edu/~keleher/bib/dsmbiblio/dsmbiblio.html
In this method, you essentially code to the pthread API and everything
happens automagically.
-Arun
-----------------------------------------------------------------------
The LIH mailing list archives are available at:
http://lists.linux-india.org/cgi-bin/wilma/linux-india-help
-----------------------------------------------------------------------
The LIH mailing list archives are available at:
http://lists.linux-india.org/cgi-bin/wilma/linux-india-help
