Homework #1

Due on July 12 (Monday)

This homework is to run parallel matrix multiplication program.

1. On the PPLab's 4-cluster, what will be the best algorithm to solve a 1000x1000 floating point matrix multiplication using MPI APIs.
2. Do the same thing using FM API's.
3. On the Cray T3E in Supercomputer Center, do the same thing on 4 nodes with MPI and compare the result.
4. Do the same thing on different number of nodes with MPI.

Assume that the value of (i,j)th element of matrix A is "i/10-j/100" and that of (i,j)th element of matrix B is "i/100-j/10". Also assume that each node has only the proportional amount of elements in its local memory. When it needs other elements, they should be sent.

* Information on PPLabs' PC Cluster

 

• Configuration : 4 nodes (Pentium II PC's)
• Location : {cluster1, cluster2, cluster3, cluster4}.icu.ac.kr
• User ID and the default password

Name	ID / Default password	Email
°­¹®¼ö	kkamo / 99kkamo	kkamo@icu.ac.kr
±è½ÅÀÏ	god4u / 99god4u	god4u@icu.ac.kr
³ª»ó¿Á	nso5540 / nso554099	nso@icu.ac.kr
¸ð»ó¸¸	smmoh / 99smmoh	smmoh@icu.ac.kr
¹Ú¿ë¹é	nausica / nausica99	nausica@icu.ac.kr
¼Õµ¿È¯	dhson / 99dhson	dhson@icu.ac.kr
¿ì°æÅÂ	hall00 / 99hall00	hall00@icu.ac.kr
À±¿µÇÏ	toreit / toreit99	toreit@icu.ac.kr
À̹αÔ	niklaus / niklaus99	niklaus@icu.ac.kr
À̽ÂÀÏ	gaialee / gaialee99	gaialee@icu.ac.kr
Àü¼ºÀÍ	sijun / 99sujun	sijun@icu.ac.kr

• How to use MPI (Message Passing Interface) on the cluster ?
• Edit your parallel program using MPI APIs
• Compile and link your program with a makefile (refer "mpi_makefile"in your home directory)
• Execute the command "your-prog -np 4 -key value xxx" on four machines (if you run your program on 4 nodes)
• See "/hpvm/Myrinet/examples" for examples
• Refer Argonne National Lab's MPI site for complete discussion on MPI as well as APIs
• Refer Peter Pacheco's tutorial on MPI

• How to use FM (Fast Message) on the cluster ?
• Edit your parallel program using FM APIs
• Compile and link your program with a makefile (refer "fm_makefile"in your home directory)
• Execute the command "your-prog -np 4 -key value xxx" on every machine you want to run (if you run your program on 4 nodes)
• See "/hpvm/Myrinet/examples" for examples
• Refer "http://www-csag.ucsd.edu/projects/hpvm/doc/hpvmdoc_42.html#SEC48" that is the subpage of UIUC's HPVM (High Performance Virtual Machine) Homepage

* Information on algorithms of the parallel matrix multiplication

• For algorithms 1 to 4 in the class, refer Tanenbaums group's Prof. Bal's Note and his class homepage at Parallel Programming
• For One-dim. algorithms, refer Prof. Demmel's Note and his class homepage at Parallel Computation
• For BLAS algorithms, refer Prof. Demmel's Note and his class homepage at Parallel Computation