CSE610 - Parallel Computer Architecture - Spring 2005 v2 25Jan05

Lectures: 1441 Computer Science   Mon, Wed 15.50-17:10 (3:50-5:10 pm)

Professor: Larry Wittie   Office meetings : 1308 Comp Sci Bldg
Email:
ldw@cs.sunysb.edu   Office hours: MW 17:30-18:15pm (maybe later also) (1308 CSB)
Phones:
(Messages only)2-8456    (NetLab 1308)2-8750    Web: http://www.cs.sunysb.edu/~cse610

Required Text (price estimates from web 1/23/05): "In Search of Clusters" 2nd Edition, by Gregory Pfister, Prentice Hall, 1998 $45 ($25 used at http://gettextbooks.com/search/?isbn=0-13-899709-8) ISBN 0-13-899709-8 *****5 of 5

Recommended Text 1: "Computer Architecture: A Quantitative Approach" THIRD Edition, by John Hennessy and David Patterson, Morgan-Kaufman, 2002 $90 ($40 used at http://gettextbooks.com/search/?isbn=1-55860-596-7) ISBN 1-55860-596-7***3.4 of 5

Recommended Text 2: "Parallel Computer Architecture: A Hardware/Software Approach", by David Culler, J.P. Singh and Anoop Gupta, Morgan-Kaufman, 1998 $90 ($40 used at http://gettextbooks.com/search/?isbn=1-55860-343-3) ISBN 1-55860-343-3 ****4.3 of 5

Objectives: This course is a lecture survey of the major objectives, constraints, and techniques for creating effective parallel computing systems. This semester we will first read the Pfister text on the power and limitations of grid machines (distributed computing).  The material in chapters 6 (Multicomputers) and 8 (Networks) of the Hennessy and Patterson classic will be supplemented by historical material from the Culler text, which is covers mainly distributed shared memory parallel computing, culminating in the SGI Origin 2000.  Material on the Cray MTA-2 (nee Tera MTA-1), the leading example of multithreaded architectural support for shared memory supercomputing will be round out the coverage of existing and historical parallel computing systems.  Most of you already own Hennessy and Patterson.  I have a few copies of Culler that can be shared.

Explicitly parallel computing historically has been the answer to the next increase in computing power for each generation of technology, only to become an invisible part of future generations. The 1960s-70s had vector pipeline supercomputers. The 70s-80s, parallel machines of many shapes to minimize switch delays for data sharing. The 80s-90s, shared memory bus multiprocessors with scalar pipeline RISC mCPUs. The 90s-00s, non-shared memory grid computers from racks of commodity (RISC and shared bus) mCPUs connected by fast network switches. What is next? Muliprocessor chips. Shared-memory grid machines spanning the globe. Invisible nanotech computing grids in wall paper, with short-distance wireless links to global grid optical fibers. Unobservable parallelism via quantum computing. What else?

Central themes will be how each architecture seeks to execute certain classes of programs faster than a single processor, the difficulty in accessing memory fast enough to keep the CPUs busy, the de-emphasis of network topologies as propagation delays dominate switch delays in both ultra-fast and Earth-grid computers, theoretical limits on parallel program speedup, and memory sharing via caches for fewer than 500 CPUs and multi-threading for massively parallel computers. Roughly 30% of the course material will come from the Pfister essay, 30% of the course material will come from the Hennessy and Patterson text, 20% of the course material will come from the Culler text, mainly chapters 1, 8, and10; the rest from newer articles and my own research.

Credits: 3

Lectures . . . . Assignments (most recent first, updated 1/25/05)
Projects (updated 0/0/00) . . . . Syllabus (updated 1/24/05)

NOTICES (most recent first)
====================================================
The class has moved to 1441 Computer Science.   First homework is on the web in Lect01 and Assignments.  The grading weights have been slightly altered in Syllabus.
====================================================

The class will be moved to a room in Computer Science.  Check this website Monday 1/24/05 early afternoon. ======================================================================================

Whether registered or not, if you want to take cse610 this semester:

  1. Come to the next class meeting !!
  2. Send me a short email telling me your:

name, email address, dept, status at StonyBrook (example, CSE 2nd yr grad), registration status for CSE610 (example, NOT Registered)

when and where you took a computer architecture course (like our cse502 or cse320 or cse345) covering instruction pipelines & memory caches, what course, your grade, max possible grade

Any comments you wish to make about your preparation and expectations for this course.

  1. Start Reading the Pfister text.

====================================================

All CSE610 Web Files

 

=============