CSE 8993

Advanced Scientific Computing

Prof. Ioana Banicescu.

HOURS:

CLASS: Tue / Thu 12:30 - 1:45, Butler 103.

OFFICE: 320 Butler Hall, Tue / Thu 3:00 - 4:00. Other times by appointment only. Email: ioana@cs.msstate.edu.

325-2756.

TEXT BOOKS:

H. El-Rewini, T. G. Lewis, H. Ali, Task Scheduling in Parallel and Distributed Systems , Prentice Hall, 1994.

C. Xu and F. Lau, Kluwer, Load Balancing in Parallel Computers - Theory and Practice , Academic Publishers, 1997.

PREREQUISITES:

A graduate course in high performance computing is a must. Graduate level courses in algorithms, parallel algorithms, operating systems, computer architecture and experience with MPI are desirable.

OBJECTIVES:

This course evolves out of the Designing Parallel Algorithms and a series of other high performance computing courses and related topics taught at the Mississippi State University. It is targeted towards graduate students in sciences and engineering who are interested in solving computationally intensive problems on parallel and distributed machines.
Since students come from different departments and backgrounds, the lectures will attempt to emphasize an even spread of topics in computer architecture, software, numerical methods, algorithm design. Students will spend time on particular parallel algorithms for scientific problems and their implementations on parallel and distributed machines. Among sample problems we can enumerate: linear algebra and large systems of differential equations arising in global climate modeling, CFD applications, hierarchical methods for particle simulations (N-body simulations for gravitational problems, molecular dynamics and plasma physics), etc.

COURSE REQUIREMENTS:

The emphasis is be on task scheduling and load balancing algorithms. We will survey recent papers addressing this algorithms and will discover their impact in improving performance of scientific applications. The class will study a variety of algorithms, their applications, and tradeoffs between different solutions. There will be discussions on issues such as: parallel computer architectures (memory hierarchy, interconnection networks, latency and bandwidth, parallel I/O), and software systems, with the aim of understanding their capabilities, costs and limitations. We will analyze and evaluate the performance of various implementations by comparing results obtained from using MPI, Matlab, NESL, CILK, etc. We will draw conclusions regarding programming paradigms, and programming environments and tools for parallel and distributed computing.
We will have invited talks related to our topics given by researchers at the ERC on: numerical analysis, CFD applications and runtime environments. There will be several assignments, a project, midterm and final paper. All projects will be presented before the end of the semester. The final grade will be based on all of the above.

There will be a midterm, a final exam and a few assignments.

GRADING: Assignments and Project(s) 30%; Midterm 20%;Presentation 10%; Final exam (or paper) 40%.

ANNOUNCEMENTS:

Students are responsible for checking the announcements regularly. For important announcements, click here.

COURSE OUTLINE:

The course will cover material from the following topics:

Scientific Computing: scientific problems and their characteristics; the impact of parallel programming on performance and numerical properties.

Fundamentals of parallel and distributed computing: models of parallel computation; parallel programming paradigms, design methodology for parallel algorithms; performance analysis and evaluation.

Scheduling Parallel Tasks: optimal scheduling algorithms; NP- completeness of the scheduling problem; considering communication delay; list scheduling heuristics; advanced scheduling heuristics; dynamic task scheduling; loop scheduling.

Load Balancing Algorithms: classification, models and performance metrics; nearest -neighbors algorithms; deterministic and stochastic algorithms; diffusion, dimension exchange and the gradient models; termination detection problem, edge-coloring, and optimality analysis (lower bounds); combinatorial optimizations (branch-and-bounds methods);

Applications: parallel unstructured grid partitioning: flow calculation, selection of vertices for load migration, experimental results; the use of recent technology.

ACADEMIC HONESTY:

Students are expected to follow the CS department's policy on academic honesty with regards to homework, projects and exams.

[INSTRUCTOR] [HOURS] [PREREQUISITES] [OBJECTIVES] [REQUIREMENTS] [GRADING] [ANNOUNCEMENTS] [OUTLINE]

[ACADEMIC HONESTY][RETURN]