Syllabus for |
|
EDA280 - Parallel computer systems |
|
Owner: TDATA |
|
4,0 Credits (ECTS 6) |
Grading: TH - Five, Four, Three, Not passed |
Level: A |
Department: 37 - COMPUTER SCIENCE AND ENGINEERING
|
Teaching language: Swedish
Course module |
|
Credit distribution |
|
Examination dates |
Sp1 |
Sp2 |
Sp3 |
Sp4 |
|
No Sp |
0193 |
Examination |
4,0 c |
Grading: TH |
|
|
|
2,0 c
|
2,0 c
|
|
|
|
20 May 2006 pm V, |
28 Aug 2006 am V |
In programs
TELTA ELECTRICAL ENGINEERING, Year 4 (elective)
TITEA SOFTWARE ENGINEERING, Year 4 (elective)
TITEA SOFTWARE ENGINEERING, Year 3 (elective)
TDATA COMPUTER SCIENCE AND ENGINEERING, Year 3 (elective)
TDATA COMPUTER SCIENCE AND ENGINEERING - Embedded computer systems engineering, Year 4 (elective)
DCMAS MSc PROGR IN DEPENDABLE COMPUTER SYSTEMS - Dependable Programming, Year 1 (elective)
DCMAS MSc PROGR IN DEPENDABLE COMPUTER SYSTEMS - Dependable Architectures, Year 1
Examiner:
Professor
Per Stenström
Eligibility:
For single subject courses within Chalmers programmes the same eligibility requirements apply, as to the programme(s) that the course is part of.
Course specific prerequisites
EDA 111 Computer Architecture
Aim
To provide a fundamental insight into the design principles for general-pupose parallel computer systems.
Goal
After completion of the course, the student will have ained conceptual as well as practical insights into prevailing parallel programming models and techniques for parallel computer systems - multiprocessors and clusters. In addition, the course will provide the student with indepth knowledge about design principles of parallel computer systems. This includes, but is not limited to, semantic models of memory systems for parallel computer systems, design principles of cache coherence and message passing protocols, interconnection networks, performance
metrics and scalability models.
Content
Shared address-space as well as distributed memory (private address-space) machines use commodity high-performance microprocessors but differ in the fundamental programming models offered to the software systems. The course first introduces the two programming models they offer to the software: The shared address-space model and the message-passing model. In laboratory work, the course participants will have a chance to get insights into parallel programming through basic programming assignments. In particular, we will learn about strategies for how algorithms that were originally developed for single-processor systems can be converted to run efficiently on parallel computers using the shared address-space programming model.
The main part of the course is devoted to design principles for parallel computer architectures. Studies will be done on how memory systems for multiprocessor and distributed memory systems are designed. The most critical aspect of this is how to deal with scalability, namely, how to make performance scale for memory systems and interconnection networks in a cost-effective manner. Important concepts that are treated are memory consistency models, cache coherence, interconnection networks, and latency tolerating techniques to mention a few.
Organisation
The course is partly a self study course where the text book defines the content. To make it easier to go through and comprehend the material, the course contains lectures, exercises and laborations. The content of the course follows closely the text book and each lecture introduces and summarizes one subject area from the text book. In total, there are 12 lectures and exercises, and 2 laborations.
Literature
Culler, Singh, Gupta. Parallel Computer Architecture: A Hardware/Software Approach. Morgan Kaufmann Publishers Inc. San Francisco, 1998.
Examination
Written exam and pass on Laboratory.