[iris-bulk] Pre-IRIS Workshop Short Course: Introduction to Multinode Computing using MPI

[IRIS]

Pre-IRIS Workshop Short Course - June 3, 2008
Introduction to Multinode Computing using MPI (Message Passing  
Interface)

IRIS-DMS will sponsor a one-day tutorial on how to utilize multi-node  
parallel computing by using the MPI (Message Passing Interface)  
library within your geophysical software codes.  Many of us have  
access to in-house or campus multi-processor computers.  However, in  
order to access many computing nodes at one time, particularly those  
distributed across a cluster, our software codes need to be programmed  
to ask for and to compute on the many nodes.  Currently the most  
commonly used method to perform this task is MPI, a set of library  
calls that performs the communication to the multiple processors.  MPI  
enables an application code to access from two to 512+ nodes (desktops  
to national supercomputer sites).
The full-day pre-workshop tutorial will provide an overview of multi- 
node parallel computing strategies including one-task-many times  
(e.g., thousands of cross-correlations) and one-task parsed and  
distributed across many processors (e.g., 3D finite difference  
algorithm).  Future directions of parallel computing will also be  
discussed.  MPI basics will be presented.  During the tutorial,  
participants will gain hands-on experience of writing simple codes  
containing MPI and then on how to run these codes.  Tutorial  
instructors will provide access to a multi-node facility at their  
national computing centers; participants should be familiar with UNIX/ 
Linux and either C or Fortran.

Date:  Tuesday, June 3; 9 am -5 pm.  DMS will provide one night of  
lodging (June 2) and breakfast/lunch will be provided.  More  
information can be obtained from David Okaya (okaya at usc.edu) or Mari  
Francissen (mari at iris.washington.edu).

http://www.iris.edu/workshops/2008/mpi/