CSI 702 Fall 2007

Dr. John Wallin

Information for Fall 2007

• PDF version of Syllabus
• Information about Computing in COS
• FAQ for COS computing
• Spring 2007 CSI 702 course
• Spring 2006 CSI 702 course
• Final from Spring 2007

Homework

• Due October 24, 2007 MPI Programming Assignment
  • Do any TWO exercises from the list.
  • Place the final writeup so it can be accessed from http://www.cos.gmu.edu/~youraccount/c702f07/mpi_hw1
• Due October 15, 2007 - NOTE NEW DEADLINE
  • Develop a simple scientific program that uses sockets and forks to distribute the work involved in the problem.
  • Write a one page summary of a computational project you want to do using High Performance Computing. This project can be based on new or existing codes where you need more runs, higher resolution, or greater speed.. Explain the basic problem and why it is important, what runs you need to have completed, an estimate of the CPU time and disk space needed for the runs, and why HPC is needed to solve it. The purpose of this paper is to give you a chance to state what types of problems you are interested in, and may serve to help guide the content of the course and the direction of future projects within the class. It also will start you thinking about concrete problem instead of just abstract concepts about computing.
  • This program can redo elements of the threads assignment, implemented with the sockets and forks.
  • Place the solution so it is accessible at http://www.cos.gmu.edu/~youraccount/c702f07/sockets
• Due September 24, 2007
  • Get an account working on the COS cluster
  • Go through the threads tutorial - including compiling and running the sample codes
  • Create an application where threads are used in scientific programming. Outline the code and explain how threads will speed up or help the execution of this application. This need not be highly complex, but it should should solve a problem.
• Due September 24, 2007
  • Develop a simple scientific program that uses the pthreads library to distribute the work involved in the problem.
  • Write a one page summary of a computational project you want to do using High Performance Computing. This project can be based on new or existing codes where you need more runs, higher resolution, or greater speed.. Explain the basic problem and why it is important, what runs you need to have completed, an estimate of the CPU time and disk space needed for the runs, and why HPC is needed to solve it. The purpose of this paper is to give you a chance to state what types of problems you are interested in, and may serve to help guide the content of the course and the direction of future projects within the class. It also will start you thinking about concrete problem instead of just abstract concepts about computing.
• Due September 17, 2007
  Suboptimal Assignment
  This assignment is designed to help you learn how optimization and "de-optimization" can affect the performance of a code. The goal is to case cache misses, breaks in the pipelines, and general wreck mayhem with all the things that make your code run fast on modern computers.

Lecture Notes

• Lecture 1 - August 29, 2007
  • Lecture Notes 1 - Wallin
• Lecture 2 - September 5, 2007
  • Lecture Notes 2 - Wallin
• Lecture 3 - September 12, 2007
  • Lecture Notes 3 - Wallin
  • Example codes for threading - download and compile using "gcc -lpthread name"
    • Threads Example Code 1 - trivial and unpredictable first program
    • Threads Example Code 2 - basic joining of threads with possible race condition on a counter
    • Threads Example Code 3 - preventing race conditions with mutex and join
    • Threads Example Code 4 - synchronizing order with mutex
    • Threads Example Code 5 - bad implementation using condition variables
    • Threads Example Code 6 - synchronizing order with condition variables
    • Threads Example Code 7 - unpredictable results from using the same variable in each thread
    • Threads Example Code 8 - an example of how things go wrong without mutex locks
• Lecture 4 - September 19, 2007
  • Lecture Notes 4 - Wallin
  • fork1.c - a very simple fork program
  • fork2.c - a fork program with sockets
  • fork3.c - forking a fork
  • fork4.c - forking a forked fork
  • semaphore_barrier.c - a simple barrier implemented with semaphores
  • client.c - a socket client implemented in C
  • client.pl - a socket client implemented in Perl
  • server.c - a socket server implemented in C
  • server.pl - a socket server implemented in Perl
• Lecture 5 - September 26, 2007
  • Lecture Notes 5 - Wallin
• Lecture 6 - October 3, 2007
  • Lecture Notes 6 - Wallin
  • Materials for Learning MPI from Argonne National Laboratory
  • Sample Fortran Code from Using MPI by by William Gropp, Ewing Lusk, and Anthony Skjellum
  • Sample C Code from Using MPI by by William Gropp, Ewing Lusk, and Anthony Skjellum
  • Sample hostfile for COS workstations
• Lecture 7 - October 10, 2007
  • Lecture Notes 7 - Wallin
  • Examples from Using MPI, 2nd Edition
• Lecture 8 - October 17, 2007
  • Lecture Notes 8 - Wallin
  • ring.c MPE "ring toss" demo code from class
  • compile_ring shell script to compile ring.c
  • compute_ring.c ring toss with computing
  • packing_test.c Demo of structure packing from class
  • mpi_put.f90 simple MPI RMA put example
  • mpi_get.f90 simple MPI RMA get example
• Lecture 9 - October 24, 2007
  • No notes are available for this lecture. It involved a case study of the mass99 code.
  • The mass99 code willbe put on this site shortly, with access only for students in the class.
• Midterm Exam - October 31, 2007
  • The exam
  • The solution
  • Broadcast Code - solution to problem 3
• Lecture 10 - November 7, 2007
  • Lecture 10 Notes
  • LLNL OpenMP Tutorial - a great resource for learning OpenMP with example codes
  • Code Examples
    • LLNL OpenMP Exercises
    • trivial.f90
    • trivial.f90
    • pi.f90 - adapted from Mattson, Sanders and McGill's book
    • mandel.f90 - a simple mandelbrot program that can be adapted to OpenMP
• Lecture 11 - November 14, 2007
  • Lecture 11 Notes
  • The GRAPE project
  • Article by Nakaasato et all on the Using reconfigurable hardware
  • Example of using graphics cards in science
  • NVIDIA Developer Site - includes tutorials and sample codes
  • CUDA NVIDIA Developer Site
  • The General-Purposing Computing Using Graphics Hardware website
• Lecture 12 - November 28, 2007
  • Lecture 12 Notes
  • Parabon Computing
  • The Teragrid
  • Teragrid Gateway Tutorial
  • Boinc volunteer computing grid

References

• IEEE Floating Point Standards by Steve Hollash at Microsoft
• MPI Forum resources
• Netlib MPI resources
• Beowulf - the history
• Beowulf - the story
• LLNL Computing Tutorials