CPS 303 High Performance Computing Fall 2011 MWF: 12:00 – 1:00 pm (Holmes 104) Instructor: Wensheng Shen Department of Computational Science 130 Smith Hall Phone: (585) 395-5182 Email: [email protected] Web: http://www.cps.brockport.edu/~shen Office hour: TR 10:00 am – 12:00 pm or by appointment Prerequisite: basic calculus, programming in C or Fortran, the UNIX operating system, linear algebra Textbook: An Introduction to Parallel Programming Author: Peter Pacheco Pulisher: Elsevier Science Ltd. ISBN: 0123742609 References: MPI - The Complete Reference Authors: Steven Huss-Lederman, William Gropp Pulisher: MIT Press Description: High-performance computing involves the use of advanced computer technology and techniques to analyze models in fields such as mathematics, engineering, business and sciences. Successful implementation of these models invariably requires the knowledge of mathematics. Frequently, the models generated can be prohibitively large (in terms of the required computational resources). Such situations require the use of parallel programming approaches. This course includes various topics in high performance computing, such as computer architectures, basic topics of parallelism, parallel computing, and understanding benchmarks, numerical integration, differentiation, linear systems and matrix-vector operations. Objectives: This course presents principles, algorithms, and strategies in high performance computing. Outcome of a successful completion of the course include: (1) to understand basic principals of parallelism, (2) to translate mathematical descriptions into models suitable for investigation via a high-level language such as variants of Fortran and C, (3) to identify the numerical techniques necessary to solve a given model, (4) to understand how to extract parallelism from numerical algorithms, (5) to implement computational algorithms using both serial and parallelprogramming techniques, (6) to gain experience in the use of message passing software (MPI), (7) to understand how to compare and evaluate the computational characteristics of computers through the use of benchmarks. Topics: The topics include Discretization of Models, Numerical Algorithms, Parallel Architectures, Concepts, Parallel Extraction, The Message Passing Interface (MPI), Benchmark Evaluation. Grading: Homework assignments (50%) Projects (30%) Midterm (10%) Final exam (10%) Average 100 – 90.0 87.0 – 89.9 83.0 – 86.9 80.0 – 82.9 77.0 – 79.9 73.0 – 76.9 Grade A A- B+ B B- C+ Average 70.0 – 72.9 67.0 – 69.9 63.0 – 66.9 60.0 – 62.9 57.0 – 59.9 <57.0 Grade C C- D+ D D- E Note: extra credits may be given in home and project assignments as well as tests. Final exam may be replaced by a project. Assignment policy: Homework assignments given in class will be due in two weeks and project assignments given in class will be due in four weeks after they are assigned. Late assignments can be accepted with a penalty at a rate of 10% per day. No makeup tests and no incompletes. A missed test will receive 0 points. Exceptions to these rules, at instructor's discretion, apply to cases of illness, personal tragedy, or extraordinary circumstances beyond a student's control, if it is documented to instructor's satisfaction. Arrangement for such an exception needs to be discussed with the instructor. Attendance: Students are expected to attend all classes. Some of the material may not be contained in the textbook. If a student misses a class, it is his/her responsibility to get class notes and handouts. Absences will be excused for documented illness, official representation of the College, an unfortunate death of a close relative, religious holiday, and other circumstances beyond student's control. Authorship: Students are allowed to discuss ideas and help others by explaining concepts and possible solutions. All the work that is submitted, however, must be performed by individual students independently. Students must provide appropriate citations for any text fragments in books, journals, conference proceedings, web-based resources, etc. that have been used in their assignments. Students also need to acknowledge any help fromothers. A student is considered cheating if he/she submits materials as his/her own work that is not entirely his/her own work, or if he/she intentionally provides an answer to another person. If cheating has been detected, the student will receive a zero grade for that assignment. Further disciplinary procedures may also be considered. Policies: Students with documented disabilities may be entitled to specific accommodations. SUNY Brockport's Office for Students with Disabilities makes this determination. Please contact the Office for Students with Disabilities at 395-5409 to inquire about obtaining an official letter to the course instructor detailing approved accommodations. The student is responsible for providing the course instructor with the official letter. Faculty and staff work as a team with the Office for Students with Disabilities to meet the needs of students with disabilitiesWeekly Schedule Week 1 Aug. 29 Class begin, syllabus, introduction Aug. 31 Chapter 1: Introduction to HPC Sept. 2 Chapter 2: Architecture of Parallel Computers Week 2 Sept. 5 Labor day Sept. 7 Chapter 2: Architecture of Parallel Computers Sept. 9 Chapter 2: Architecture of Parallel Computers Week 3 Sept. 12 Chapter 2: Architecture of Parallel Computers Sept. 14 Chapter 3: Basic MPI Sept. 16 Chapter 3: Basic MPI Week 4 Sept. 19 Chapter 3: Basic MPI Sept. 21 Chapter 4: An application --- numerical integration Sept. 23 Computer lab: Basic MPI Week 5 Sept. 26 Chapter 4: An application --- numerical integration Sept. 28 Chapter 4: An application --- numerical integration Sept. 30 Computer lab: Numerical integration Week 6 Oct. 3 Chapter 5: Collective communication Oct. 5 Chapter 5: Collective communication Oct. 7 Computer lab: Collective communication Week 7 Oct. 10 Chapter 5: Collective communication Oct. 12 Chapter 6: Grouping Data for Communication Oct. 14 Midterm exam Week 8 Oct. 17 Fall break Oct. 19 Chapter 6: Grouping Data for Communication Oct. 21 Computer lab: Grouping Data for Communication Week 9 Oct. 24 Chapter 7: Communicators and topologies Oct. 26 Chapter 7: Communicators and topologies Oct. 28 Computer lab: Communicators and topologies Week 10 Oct.