a question about MPICH

[Bohn Christopher A Capt AFRL/IFSD]

Goo day,

> From: Ahmet Tokuz [mailto:atokuz@yahoo.com]
> Sent: Sunday, June 20, 1999 7:27 PM
[...]
> we can run all the programs and benchmark tests
> seamlessly, even I
> didnt share the MPI binaries with NFS. What i did is just to put the
> binary which must be run with mpirun to shared home directory and let
> it run. It works perfectly.

Not surprising at all.  With MPICH, you static-link the libraries.  You
could, in theory, even delete MPICH after compilation.

> My question is one of our supervisors told me that our master 
> node will
> be single point of failure and it will cause bottleneck..So I am
> supposed to install MPICH to all other nodes. Does it help to get rid
> of single point of failure problem and eliminate the 
> bottleneck? Or let
> me ask in another way if I do so, will network traffic decrease?

The situation you describe does not make the master a single point of
failure as far as MPI is concerned.  It is a single point of failure as far
as your executable goes (if your NFS server goes down, you obviously cannot
access the file), and your entire cluster is multiple point of failure as
far as actual execution (it ain't pretty when a node goes down during
execution, and high availability is a different mailing list, just to cut
that off at the pass).

So, the short answer is that placing MPICH on all nodes will not decrease
network traffic.

Take care,
cb
*-*-*-*-*                                                     *-*-*-*-*
Christopher A. Bohn, Capt, USAF     || christopher.bohn@sn.wpafb.af.mil
Digital Simulation Systems Engineer || cbohn@computer.org
Collaborative Simulation Technology ||
          and Applications Branch   || v (937)255-4429x3576  (DSN785)
Information Directorate             || f (937)255-4511       (DSN785)
Wright Research Site                ||
Air Force Research Laboratory       || http://members.aol.com/EngrBohn/
         http://www.if.afrl.af.mil/div/IFS/IFSD/IFSD_home.html
*-*-*-*-*                                                     *-*-*-*-*