Programming Assignment 5
Matrix Multiplication using Hive

Due on Thursday November 10 before midnight


Description

The purpose of this project is to develop a simple program for matrix multiplication using Apache Hive.

This project must be done individually. No copying is permitted. Note: We will use a system for detecting software plagiarism, called Moss, which is an automatic system for determining the similarity of programs. That is, your program will be compared with the programs of the other students in class as well as with the programs submitted in previous years. This program will find similarities even if you rename variables, move code, change code structure, etc.

Note that, if you use a Search Engine to find similar programs on the web, we will find these programs too. So don't do it because you will get caught and you will get an F in the course (this is cheating). Don't look for code to use for your project on the web or from other students (current or past). Just do your project alone using the help given in this project description and from your instructor and GTA only.

Platform

As in the previous projects, you will develop your program on SDSC Comet.

Setting up your Project

Login into Comet and download and untar project5:

wget http://lambda.uta.edu/cse6331/project5.tgz
tar xfz project5.tgz
chmod -R g-wrx,o-wrx project5
You may use Hive on Comet in local mode interactively, but you need to setup your PATH (you need to do this every time you login to comet):
source ~/project5/setup
You also need to create an empty metastore database first (this must be done only once):
cd
schematool -dbType derby -initSchema
Then, to evaluate Hive commands interactively, do:
hive
Go to project5/example and look at the join.hql example. You can run it in local mode (after you setup your PATH) using:
hive -f join.hql

Project Description

You are asked to re-implement Project #1 (matrix multiplication) using Apache Hive. This time, you need to store the result of the multiplication into a Hive table and then write a Hive query that counts the number of matrix elements and the average matrix value of the multiplication result. An empty multiply.hql is provided as well as a script to run this code on Comet. The input matrices are the same as in Project1. There are two small sparce matrices 4*3 and 3*3 in the files M-matrix-small.txt and N-matrix-small.txt for testing in local mode. For these matrices, your program should print the following COUNT and AVG:

12	15.5
Then there are 2 moderate-sized matrices 200*100 and 100*300 in the files M-matrix-large.txt and M-matrix-large.txt for testing in distributed mode. For these matrices, your program should print the following COUNT and AVG:
59889	-0.06041565604668677
Note: you can access the input matirces in Hive (which are passed as parameters) as ${hiveconf:N} and ${hiveconf:M}.

To run it in local mode over the two small matrices do:

hive -f multiply.hql --hiveconf M=M-matrix-small.txt --hiveconf N=N-matrix-small.txt
To dump the output to the file multiply.local.out, do:
hive -f multiply.hql --hiveconf M=M-matrix-small.txt --hiveconf N=N-matrix-small.txt &>multiply.local.out
After you make sure that your program runs correctly in local mode, you run it in distributed mode using:
sbatch multiply.distr.run
This will multiply the moderate-sized matrices.

Documentation

You can learn more about Hive at:

What to Submit

You need to submit the following files only:

project5/multiply.hql
project5/multiply.local.out
project5/multiply.distr.out

Submit Programming Assignment #5:

Last modified: 10/26/2016 by Leonidas Fegaras