CS528-PARALLEL NUMERICAL COMPUTING

TIME: TUE- 6:10-8:50. SEC-209.

INSTRUCTOR: Apostolos Gerasoulis, gerasoul@cs.rutgers.edu, CORE-309, x2725.
OFFICE HOURS: Tu, Th 5-6 and by appointment.

TA: Zhijian Lu, zhlu@paul, Core 338, Tel: 445-5373.

SUMMARY: 

A. AN AD HOC INTRODUCTION TO PARALLEL PROGRAMMING
      MPI programming interface
      Point to point communication, Datatypes and Packing
      Collective Communications, Communicators, Topologies.
      Examples:Matrix Multiplication	
B. PARALLEL ARCHITECTURES
   Introduction to parallelism. 
   Graph representation of parallelism and parallel architectures.
         Interconnection networks 
         Topological properties. 
	 Communication algorithms.
    The hypercube architecture. Programming the NCUBE-2.
C. PROGRAM PARTITIONING  AND PARALLELISM IDENTIFICATION
	Data dependence in loops
	Data dependence analysis
	Loop Restructuring
C. SCHEDULING.
	Static, Dynamic
	Algorithms
	granularity
	Clustering
D  TOOLS FOR PARALLEL PROCESSING.
   PlusPyr, PYRROS
E  ADVANCED APPLICATIONS
	Gaussian Elimination.
        Sparse Matrix
        NBODY computation.
	Iterative linear algebra.


WORK: 30% HW, 35% Midterm and 35% Final
      Parallel programming projects + Regular HW.

BOOKS: 
R-1.	Parallel Computing, Kumar,Grama, Gupta, Karypis.
R-2.    Class Notes and papers.
R-3. 	High Performance Compilers for Parallel Computing, M. Wolfe.
R-4     MPI books[ 1. The complete Reference 2. Using MPI