To introduce students to high performance computing techniques involving parallel computing for large scale numerical models.
The computing language used will be Fortran, the basics of which will be covered in the first part of the class. Although Fortran is by now a classical programming language, its formula-oriented philosophy still makes it a widely used language inside libraries for scientific computing.
Students will develop and test their software in the parallel computing facilities in the Math Department as well as at the Pittsburgh Supercomputing Center.
- A basic knowledge of any one of the following programming languages/environments: BASIC, C, C++, Fortran, Java, MATLAB, PASCAL.
- Linear algebra and calculus.
- One introductory computational/numerical analysis class.
- The syntax of the Fortran language will be presented in the first two weeks of the course, with examples. No previous knowledge of Fortran is expected.
- The Message Passing Interface. (See, for example, the official web page or a nice tutorial.) No previous knowledge of MPI is expected.
- Open MP. (See, for example, the official web page or a nice tutorial.) No previous knowledge of Open MP is expected.
- Elementary numerical methods for partial differential equations,
primarily the Poisson equation (
) and the heat equation
(
) in two dimensions.
- Elementary Monte Carlo methods for integration.
- Design of parallel programs.
- Measuring performance of parallel programs.
- Special attention will be paid to assuring code fidelity and debugging.
- Numerical linear algebra: solving linear equations in parallel.
- Software packages such as LAPACK and PETSc.