EECE 422 Syllabus, Fall 2013

Department of Electrical and Computer Engineering Faculty of Engineering and Architecture American University of Beirut EECE 422 – Parallel Computer Architecture and Programming Catalog description: A course on high-performance computer architectures with emphasis on shared memory and distributed parallel architectures and programming models. Topics include: simultaneous multithreading processors, multicore processors, SIMD processors, UMA, NUMA and COMA shared-memory multiprocessors, distributed multiprocessors, snoopy and directory-based cache coherence protocols, memory consistency models, high performance synchronization methods, speculative lock elision, shared memory programming model, message passing programming model and transactional memory programming model. To consolidate the material presented in class, students work on designing parallel programs using the OpenMP threading environment and MPI message passing programming standard. Credit hours: 3 credits Required or elective: Elective for ECE and CCE students Prerequisites: By course: EECE 321, By topic: Computer Organization Textbook(s) and/or required materials: David E. Culler and Jaswinder Pal Singh, with Anoop Gupta. Parallel Computer Architecture: A Hardware/Software Approach. Morgan Kaufmann, 1998. Computer usage: OpenMP and MPI parallel programming standards Course Objectives 1. Understand how multithreaded processors work. 2. Understand how SIMD processors work. 3. Understand how multicore and multiprocessor computer systems work. 4. Understand how multiprocessors implement advanced synchronization techniques 5. Learn the identifying characteristics of popular parallel programming models. 6. Learn how to design parallel programs using OpenMP and MPI. Course Topics 1. Introduction 2. Multithreaded processors 3. SIMD/Vector Processors 4. Multicore and Uniform Memory (UMA) Access processors and Snoopy Cache Coherence 5. Distributed Multiprocessors and Directory Based Cache Coherence 6. Cache Only Memory Architecture (COMA) 7. Memory Consistency models 8. Synchronization with locks in shared memory multiprocessors 9. Speculative Lock Elision and Transactional Memory 10. Parallel programming models 11. Designing Parallel Programs: partitioning, communication and synchronization, data dependences, granularity and load balancing, I/O, performance analysis and tuning 12. Parallel programming with OpenMP 13. Parallel programming with MPI Course Learning Outcomes 1. Explain how vector/SIMD architectures enhance execution performance. 2. Explain different types of multiprocessor architectures and multiprocessor interconnects. 3. Explain different cache coherence techniques and how they are supported in hardware. 4. Explain different memory consistency models and their role in shared memory programming. 5. Explain how synchronization with locks works in a shared memory multiprocessor system. 6. Explain benefits and hardware support of speculative lock elision and transactional memory. 7. Write parallel programs using OpenMP. 8. Write parallel programs using MPI. Class/laboratory schedule: Two 75-minutesd lectures or three 50-minute lectures per week. Evaluation methods 1. Programming Assignments 30% 2. Quizzes 40% 3. Final Exam 30% Professional component Engineering topics: 100% General education: 0% Mathematics and basic sciences: 0% Person(s) who prepared this description and date of preparation Haitham Akkary, October 2012. Date of last revision: Feb 3, 2013.