Course:
Term:
Format:
Computers:
Instructor:
Phone:
Email:
Textbook:
CSCI 490 Special Topics: Introduction to Parallel Programming
Full Summer 2006
Online (Web)
Remote login to MCSR supercomputers for assignments
Jason Hale
662-915-3922
jghale@olemiss.edu
Description:
Introduction to Parallel Programming
Prerequisites:
Experience programming in Java, Fortran, or C/C++ (Preferred).
"Parallel Programming: Techniques and Applications Using Networked Workstations
and Parallel Computers" by Barry Wilkinson and Michael Allen. Prentice Hall.
http://vig.prenhall.com/catalog/academic/product/0,1144,0131405632,00.html
Course Description:
This course will introduce students to the rationale, concepts, and techniques for writing parallel
computer programs. Students will be given accounts on a supercomputer and cluster computer at the
Mississippi Center for Supercomputing Research (http://www.mcsr.olemiss.edu), and will use these
accounts to complete five programming assignments. Distributed memory (message passing)
techniques will be featured in some of the assignments. Shared memory techniques (threads) will be
featured in others. In addition to learning the programming techniques for sending messages between
processes, and for coordinating access to shared memory locations between threads, students will
learn how to think about restructuring a computing problem in such a way as to keep all available
processors gainfully employed in collectively reaching a solution as fast as possible. Students will learn
to evaluate the potential speedup of a computing problem, as well as how to measure the actual
speedup of a problem on a given architecture, and to measure how efficient the computer is in solving
the problem.
This will be a somewhat self-paced course.
Course Topics:
Introduction to Linux/Unix O/S environment
Rationale for Parallel Computers; Grand Challenge Problems
Estimating and Measuring Speedup of a Program on a Supercomputer
Measuring the Efficiency of a Parallel Computation on a Supercomputer
Design Tradeoffs of Major Classes of Parallel Computing Architectures
Submitting Batch Jobs to a Scheduler on a Supercomputer (using PBS)
Static vs. Dynamic Process Creation
Single Program Multiple Data (SPMD) vs. Multiple Program Multiple Data (MPMD) Approaches
Message Passing (MPI) Techniques and Concepts:
Communicators, Tags, Blocking vs. Non-blocking Messages, Collective Communication
Constructs (Reduce, Gather, Broadcast, Scatter)
Shared Memory Computing via Unix/Linux Heavyweight Processes
Shared Memory Computing via Unix/Linux Lightweight Threads
OpenMP Message Passing Interface for Creating Shared Memory Programs
Static Work Partitioning
Dynamic Work Allocation/Load Balancing
Grading Policy:
5 Programs (100 points each):
Several Homework Assignments:
Two Tests: (100 points each):
500 points
300 points
200 points