This course exposes students to the principal issues involved in software development for parallel computing and discusses a number of approaches to handle the problems and opportunities caused by the increased availability of parallel platforms.
The course includes lectures, assignments, self-study, and a project. 50% of your grade is determined by project work and 50% is determined by a written exam; the exam is given during the official examination period, and there is no makeup exam. Students must be able to program using Java and C/C++.
The course may cover: memory coherence and consistency models, implications for language-specific memory models, Java memory model, models of parallel programming and parallel program execution, performance models for parallel systems, tranactional memory, compiler extraction of parallelism, language and compiler support for parallel programming, threads and their execution environment, synchronization, and implementation issues of these topics.
Recitation sessions take place Thursdays 13:15 -- 15:00 in IFW D42 and take place when announced. Some of the lecture hours will be devoted to other activities (tutorials, reviews, etc) or will be devoted to group meetings. Please watch this page for updates and announcements.
This is a plan. No plan survives contact with reality.
|0||-||09/20: No meeting|
|1||09/24: Admin issues, Introduction, Parallel computing landscape Cache coherence||09/27: Cache coherence, continued|
|2||10/01: Memory consistency (Notes)||10/04: Memory consistency (cont'd), homework discussion|
|3||10/08: Reasoning about performance: PRAM model (Amdahl's law slides)||10/11: No recitation|
|4||10/15: Locks (Notes (not checked, use at own risk))||10/18: Project ideas|
|5||10/22: Locks and Lock-free (Notes (not checked, use at own risk))||10/25:|
|6||10/27: Group communication (Notes (not checked, use at own risk))||11/01: Discussion of homework|
|7||11/05: Project proposals||11/08: Discussion of homework|
|8||11/12: Reasoning about performance: Little's law, roofline model, balance principles||11/15: --|
|9||11/19: Greedy and work stealing scheduler Scheduling notes Linearizability||11/22: --|
|10||11/26: Linearizability continued Java memory model||11/29: Project progress presentations|
|11||12/03: Guest Lecture||12/06: --|
|12||12/10: Transactional Memory||12/13: --|
|13||12/17: Student project presentations|
|Thursday 13:15 - 15:00|
|Room||Assistant||nethz ID||IFW D42||Albert Noll / Zoltán Majó|
Assignments are an important part of the course. You will not learn this material from listening to a lecture alone -- you have to do the assignments.
Note: Do not hesitate to write on the mailing list or make an appointment with your TA if you have trouble with the assignments!
Any code fragments are distributed via subversion. Your solutions to the assignments must be submitted via subversion as well. Once you have registered a team, a directory will be made for you. The URL for your team is: https://svn.inf.ethz.ch/svn/trg/ps_students/trunk/2012hs/teams/<YourTeam>. You should log in with your nethz log in and password. If there are any problems please contact a TA immediately.
|2||1||09/28||10/04||Homework 1||Solution Homework 1|
|3||2||10/09||10/18||Homework 2||Solution Homework 2|
|4||3||10/18||10/25||Homework 3||Solution Homework 3|
|5||4||10/27||11/01||Homework 4||Recitation Slides array_queue_lock.c array_queue_lock_opt.c mcs_lock.c|
|7||5||11/01||11/08||Homework 5||Solution Homework 5|
|12||6||12/07||12/20||Homework 6||Solution Homework 6-->|
Every project has a man advisor that can be contacted for questions or feedback (TG = Thomas Gross, TH = Torsten Hoefler, MP = Markus Püschel, ZM = Zoltán Majó)
50% of your grade is determined by the project, and the other 50% of the grade is determined by a written 2 hr exam.