Parallel Computing Made Easy with STAR-P

February 24, 2004

SIAM Conference on Parallel Processing for Scientific Computing (PP04)

Short Course Organizer

Alan Edelman
Massachusetts Institute of Technology
[email protected]

Rationale

For years we have lived with parallel computing environments that are too low level. This tutorial will enable scientists and engineers to be more productive and solve large problems on high performance computers and clusters by introducing the parallel STAR-P environment.

Lecturers

Alan Edelman, Professor, MIT
Professor Edelman has seen the ups and downs of parallel computing for nearly two decades. He believes that supercomputers can be made easier to use. His awards include a Gordon Bell Prize for supercomputing. Edelman has worked for IBM, Thinking Machines, Akamai, SGI, Pixar, CERFACS and Lawrence Berkeley Laboratory.

Ron Choy, PhD candidate, MIT
Ron Choy's research includes the development of STAR-P version 2 and its applications. His MS dissertation, "MATLAB*P 2.0, Interactive Supercomputing Made Practical" and his parallel MATLAB survey are widely recognized in the field.

John Gilbert, Professor, University of California, Santa Barbara
Professor Gilbert was an architect of MATLAB's sparse matrix support. He has worked in parallel computation, mathematical software, and numerical and discrete algorithms. Before coming to UCSB he was an associate professor at Cornell and a principal scientist and research manager at Xerox PARC.

Viral Shah, PhD Candidate, University of California, Santa Barbara
Viral Shah's research includes the development of sparse matrix functionality for STAR-P and parallel sparse matrix methods.

Description

STAR-P takes the key features of MATLAB functions and shifts the model to the parallel universe. With a focus on providing an easy-to-use, interactive environment for high performance computing, STAR-P applies the design principles of MATLAB in a supercomputing environment. STAR-P is an ongoing research project between MIT and UC Santa Barbara.

Course Objectives

It is our hope that we will be able to:
1) Reduce the programming time for new applications
2) Reuse and leverage existing code

Level of Material

The level of material in this course will range from introductory to advanced. Some familiarity with the MATLAB environment is assumed. However, if you are not familiar with MATLAB, this will not hinder your experience in any way.

Who Should Attend

Anyone interested in parallel computation should attend. The tutorial is valuable for anyone who has MATLAB programs, and wishes to take the plunge into parallelism. This tutorial is also appropriate for those who write parallel programs, and would like to increase their productivity by programming in a higher level language.

Recommended Background

Some familiarity with MATLAB is useful but not essential. Basic background in parallel computation will not hurt, but is not necessary either.

Course Outline

1. Challenges in the parallel computing environment.
   
   Purpose of STAR-P.
   
2. STAR-P in a Nutshell
   
   Introduction.
   "Hello, World" in STAR-P.
   Philosophy behind STAR-P.
   Leveraging existing parallel libraries and more.
   STAR-P in the mainstream.
   
3. Key Features of STAR-P
   
   Starting and quitting.
   Basic Commands.
   Running small programs.
   Matrix Constructors.
   Input/Output.
   Performance tuning.
   Running on Linux and UNIX clusters.
   STAR-P on the grid.
   
4. Sample Applications
   
   Simple version of PageRank.
   Mesh Partitioning.