Tutorial: C + MySQL + MPI
--D. Thiebaut (talk) 16:30, 13 October 2013 (EDT)
The purpose of this tutorial/lab is to generate an MPI program written in C that walks a directory containing image files, gets their geometry in the form of a width and height, and enters this information in a MySQL database.
Contents
References
Verify that MPI works
- test your installation with the classic hello world program.
Source
// mpi_hello.c
#include <mpi.h>
#include <stdio.h>
int main (int argc, char* argv[])
{
int rank, size;
MPI_Init (&argc, &argv); /* starts MPI */
MPI_Comm_rank (MPI_COMM_WORLD, &rank); /* get current process id */
MPI_Comm_size (MPI_COMM_WORLD, &size); /* get number of processes */
printf( "Hello world from process %d of %d\n", rank, size );
MPI_Finalize();
return 0;
}
Compile & Run
mpicc -o hello mpi_hello.c mpirun -np 2 ./hello Hello world from process 0 of 2 Hello world from process 1 of 2
Verify that the MySQL API works
Using the example and tricks provided at http://www.cyberciti.biz/tips/linux-unix-connect-mysql-c-api-program.html, we can easily test whether we can access a MySQL database from our program:
Source
// mysqlTest.c
// Taken from http://www.cyberciti.biz
// Lists all the tables found in a MySQL database whose name is stored
// in the char[] database variable.
#include <mysql.h>
#include <stdio.h>
#include <stdlib.h>
main() {
MYSQL *conn;
MYSQL_RES *res;
MYSQL_ROW row;
char *server = "localhost";
char *user = "352a";
char *password = "xxxxxx";
char *database = "enwiki_images";
conn = mysql_init(NULL);
/* Connect to database */
if (!mysql_real_connect(conn, server,
user, password, database, 0, NULL, 0)) {
fprintf(stderr, "%s\n", mysql_error(conn));
exit(1);
}
/* send SQL query */
if (mysql_query(conn, "show tables")) {
fprintf(stderr, "%s\n", mysql_error(conn));
exit(1);
}
res = mysql_use_result(conn);
/* output table name */
printf("MySQL Tables in mysql database:\n");
while ((row = mysql_fetch_row(res)) != NULL)
printf("%s \n", row[0]);
/* close connection */
mysql_free_result(res);
mysql_close(conn);
}
Compile & Run
- Once you have installed mysql_config on your system, you can easily get the library switches and compiler switches corresponding to your installation. They are given by
- mysql_config --cflags
- mysql_config --libs
mysql_config --cflags -I/usr/include/mysql -DBIG_JOINS=1 -fno-strict-aliasing -g -DNDEBUG mysql_config --libs -L/usr/lib/x86_64-linux-gnu -lmysqlclient -lpthread -lz -lm -ldl
To compile, simply pass on the output of the mysql_config commands, or substitute their output in the compile line:
gcc -o mysqlTest $(mysql_config --cflags) mysqlTest.c $(mysql_config --libs)
To run:
./mysqlTest images <--- the only table was images
Calling System Applications from C
In our application we need to process a large collection of images and get their geometry information. To get the width and height of images, we simply call imageMagick from the C program and grab the width and height from the command.
Source
// getImageInfo.c
// D. Thiebaut
// Syntax:
// getImageInfo imageFileName
//
// Takes the image file and get identify (part of the ImageMagick tools) to
// get the image width and height in pixels.
//
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
int main( int argc, char *argv[] ) {
FILE* fp;
char command[100];
char buffer[1000];
int width, height;
//--- display syntax info if user does not specify image name---
if ( argc <= 1 ) {
fprintf( stderr, "Syntax: getImageInfo imageFilename\n\n" );
exit( 1 );
}
//--- create the command with the file name (passed in argv) ---
strcpy( command, "/usr/bin/identify -format \"%w %h\" " );
strcat( command, argv[1] );
//--- open a pipe, make the command run, and return the information ---
//--- in the pipe. ---
if ( ( fp = popen( command, "r" ) ) != NULL ) {
while ( fgets( buffer, 1000, fp ) != NULL ) {
printf( "%s", buffer );
char *p = buffer;
while ( *p != ' ' ) p++;
*p = '\0';
//printf( "width = %s height = %s\n", buffer, p );
width = atoi( buffer );
height = atoi( p+1 );
printf( "width = %d height = %d\n", width, height );
}
}
//--- close pipe ---
pclose( fp );
return 0;
}
Compile & Run
gcc -o getImageInfo getImageInfo.c ./getImageInfo bunny.jpg 1280 800 width = 1280 height = 800
Walking a Directory in C
This program is for running through our tree of directories containing images and to take all the images in one directory and
process them.
Source
#include <stdio.h>
#include <stdlib.h>
#include <dirent.h>
int main( int argc, char **argv) {
DIR *dir;
char path[] = "/media/dominique/3TB/mediawiki/images/wikipedia/en/0/01";
int count = 0;
struct dirent *ent;
if ( argc < 2 ) {
fprintf( stderr, "Syntax: ./walk path" );
exit( 0 );
}
if ((dir = opendir ( path )) != NULL) {
//-- print all the files and directories within directory
while ((ent = readdir (dir)) != NULL) {
printf ( "%s\n", ent->d_name);
//if ( count++ > 10 )
// break;
}
closedir (dir);
return( 0 );
}
else {
/* could not open directory */
perror ("");
exit( 1 );
}
}
Compile and Run
gcc -o walk walk.c ./walk /media/dominique/3TB/mediawiki/images/wikipedia/en/0/01 . .. Keep_the_Home_Fires_Burning_-_Frederick_Wheeler.ogg Voyage_of_the_Jerle_Shannara.jpg Picw.jpg ... The_Constructicons.jpg StarFeathers.jpg Brickell_district,_Miami.jpg One_Tree_Hill_-_Season_2_-_DVD.JPG Silent_All_These_Years_(Single).png
Serial Walk-Images-Store-Database Version
The version below is a serial version that walks the directory and enters the image file names into the database.
Source Program
// walkGetSizeAddMySQL.c
// D. Thiebaut
//
// See http://cs.smith.edu/dftwiki/index.php?title=Tutorial:_C_%2B_MySQL_%2B_MPI for
// more information.
//
// to Compile and Run:
// gcc -o walkGetSizeAddMySQL $(mysql_config --cflags) walkGetSizeAddMySQL.c $(mysql_config --libs)
// ./walkGetSizeAddMySQL path n/nn
//
// where "path" is the path to the root directory containing subdirectories of the form 0/00, 0/01, 0/02... a/af, each one
// containing a flat list of image files. The n/nn is one of the subdirectories, for example 0/0a.
//
#include <stdio.h>
#include <stdlib.h>
#include <dirent.h>
#include <string.h>
#include <mysql.h>
//--- a simple structure to create a pair of ints ---
typedef struct {
int width;
int height;
} Pair;
//--- Globals refering to the MySql database ---
MYSQL *conn;
MYSQL_RES *res;
MYSQL_ROW row;
char *server = "localhost";
char *user = "352a";
char *password = "xxxxxxx"; // MySql password
char *database = "enwiki_images";
int debug = 0; // use to debug the application
//-----------------------------------------------------
// getPrintTableNames: opens the database, and gets the
// names of the table(s) located in the database
// Taken from
// http://www.cyberciti.biz/tips/linux-unix-connect-mysql-c-api-program.html
//
void getPrintTableNames() {
conn = mysql_init(NULL);
/* Connect to database */
if (!mysql_real_connect(conn, server,
user, password, database, 0, NULL, 0)) {
fprintf(stderr, "%s\n", mysql_error(conn));
exit(1);
}
/* send SQL query */
if (mysql_query(conn, "show tables")) {
fprintf(stderr, "%s\n", mysql_error(conn));
exit(1);
}
res = mysql_use_result(conn);
/* output table name */
printf("MySQL Tables in mysql database:\n");
while ((row = mysql_fetch_row(res)) != NULL)
printf("%s \n", row[0]);
/* close connection */
mysql_free_result(res);
mysql_close(conn);
}
//-----------------------------------------------------
// insertFileInDatabase: inserts the name of the image
// file, its path (as in "0/01", its width and height,
// and its scale in the database ase
void insertFileInDatabase( char* filename, char* path, int width, int height, float scale ) {
char query[200];
char escapedFileName[300];
mysql_real_escape_string( conn, escapedFileName, filename, strlen( filename ) );
sprintf( query, "INSERT INTO images2 (name, path, width, height, scale1) VALUES ( '%s', '%s', %d, %d, %f )",
escapedFileName, path, width, height, scale );
// printf( "Query = %s\n", query );
int retCode = mysql_query( conn, query );
if ( retCode ) {
fprintf( stderr, "%s\n", mysql_error( conn ) );
exit( 1 );
}
}
//-----------------------------------------------------
// openDatabase: opens a connection to the database that
// is kept in the global conn variable.
void openDatabase() {
conn = mysql_init( NULL );
/* Connect to database */
if ( !mysql_real_connect( conn, server,
user, password, database, 0, NULL, 0 ) ) {
fprintf( stderr, "%s\n", mysql_error(conn) );
exit(1);
}
}
//-----------------------------------------------------
// closeDatabase: closes the connection.
void closeDatabase() {
mysql_close( conn );
}
//-----------------------------------------------------
// getFileGeometry. Calls the system 'identify' command
// to get the geometry of the file, expressed as its
// width and height.
Pair getFileGeometry( char* fileName ) {
char command[100];
char buffer[1000];
FILE* fp;
Pair pair;
pair.width = 0;
pair.height = 0;
//--- create the command and put " around the file name ---
strcpy( command, "/usr/bin/identify -format \"%w %h\" \"" );
strcat( command, fileName );
strcat( command, "\"" );
//--- open a pipe to get the output of identify ---
if ( ( fp = popen( command, "r" ) ) != NULL ) {
while ( fgets( buffer, 1000, fp ) != NULL ) {
//--- split the width and height into two strings ---
char *p = buffer;
while ( *p != ' ' )
p++;
*p = '\0';
//--- convert them both to ints ---
pair.width = atoi( buffer );
pair.height = atoi( p+1 );
}
}
pclose( fp );
//--- return them in a structure ---
return pair;
}
//-----------------------------------------------------
// getExtension. returns a pointer to the extension of
// the image file, e.g. "jpg"
const char *getExtension(const char *filename) {
const char *dot = strrchr(filename, '.');
if ( !dot || dot == filename )
return "";
return dot + 1;
}
//-----------------------------------------------------
// escapeString: puts \ chars in front of special characters
// that might be in the image file-name, so that it can be
// passed to the "identify" command and not cause the shell
// to misunderstand the name for some shell syntax.
void escapeString( char* name, char* escapedName ) {
char *p = name, *q = escapedName;
while ( *p != '\0' ) {
if ( *p == '\'' || *p == ';' || *p == '"' || *p == '(' || *p == ')'
|| *p == ' ' || *p == ',' || *p == ':' || *p == '&' ) {
*(q++) = '\\';
}
*(q++) = *(p++);
}
*q = '\0';
}
//---------------------------------------------------------------------------
// M A I N
//---------------------------------------------------------------------------
int main( int argc, char **argv) {
DIR *dir;
char path[100];
char shortPath[10];
int count = 0;
struct dirent *ent;
Pair pair;
char pathAndFileName[1000];
char escapedFileName[1000];
int noEntries = 0;
if ( argc < 3 ) {
fprintf( stderr, "Syntax: ./walkGetSizeAddMySql path n/nn" );
fprintf( stderr, "where path is the rooted path of where the image\n" );
fprintf( stderr, "resides, and n/nn is the directory-pair that is\n" );
fprintf( stderr, "md5 hash of the file name. Typically 0/01, or a/0a" );
exit( 0 );
}
//--- put the arguments into variables ---
strcpy( path, argv[1] );
strcpy( shortPath, argv[2] );
if ( path[ strlen( path )-1 ] != '/' )
strcat( path, "/" );
strcat( path, argv[2] );
//--- open connection to database ---
openDatabase();
//--- start walking the image repository ---
if ((dir = opendir ( path )) != NULL) {
//-- go recursively ---
while ((ent = readdir (dir)) != NULL) {
char* fileName = ent->d_name;
char extension[10];
//--- get file extension ---
strcpy( extension, getExtension( fileName ) );
//--- skip . and .. file names ---
if ( !strcmp( fileName, "." ) || !strcmp( fileName, ".." ) )
continue;
//--- skip some format that "identify" cannot recognize ---
if ( !strcmp( extension, "ogg" ) ) continue;
if ( !strcmp( extension, "OGG" ) ) continue;
if ( !strcmp( extension, "svg" ) ) continue;
if ( !strcmp( extension, "SVG" ) ) continue;
if ( !strcmp( extension, "mid" ) ) continue;
if ( !strcmp( extension, "MID" ) ) continue;
//--- escape special characters for identify ---
escapeString( fileName, escapedFileName );
//--- create the full path of the image file name ---
strcpy( pathAndFileName, "\"" );
strcat( pathAndFileName, path );
strcat( pathAndFileName, "/" );
strcat( pathAndFileName, escapedFileName );
strcat( pathAndFileName, "\"" );
//--- get the geometry as read by "identify" ---
pair = getFileGeometry( pathAndFileName );
//--- keep track of how many files we process ---
noEntries++;
if ( debug )
printf ( "%s\t\t%s\t%d %d\n",getExtension( fileName ), fileName,
pair.width, pair.height );
//--- add file and its geometry to database ---
insertFileInDatabase( fileName, shortPath, pair.width, pair.height, 1.0 );
}
//--- done! Close everything ---
closedir(dir);
closeDatabase();
printf( "%d images processed\n\n", noEntries );
return( 0 );
}
//--- if we're here, there was an error reading the root directory ---
else {
/* could not open directory */
perror ("");
exit( 1 );
}
}
To Compile and Run
gcc -o walkGetSizeAddMySQL $(mysql_config --cflags) walkGetSizeAddMySQL.c $(mysql_config --libs) ./walkGetSizeAddMySQL path n/nn
The MPI Version
The parallel version of this program is written using MPI as the parallel application and can run on multi-core systems as well as a cluster of distributed machines.
The first goal of writing such a parallel application is to take care of the protocol first. A good approach is to adopt the Manager/Worker paradigm, where the manager walks the directory, passes the files in blocks to each worker, which gets the geometry for each one and inserts the tuplet of file-name, path, geometry in the MySQL database.
Challenge 1 |
Given that there will be 1 manager and several workers, the first action item is to figure out who is walking the directory, who is identifying the files with ImageMagick, and who is inserting the files and geometry in the database.
Figure out a good strategy for the roles to be played by the manager and its workers.
Challenge 2 |
Figure out what needs to be exchanged between the manager and the workers. This will be a packet of information. Define exactly how this packet will be organized. Is it an array of ints, an array of chars, and array of something else? If it's an array of chars, are the strings somehow grouped as tuples? For example, if you decide that the manager does the walking and sends a group of file names to a worker, it might have to send the path for each file. Is it send as a whole string, or as a collection of (path, short-path, filename)? What separator will you use if you are sending arrays of chars?
Will workers ever need to send something back to the manager?
Challenge 3 |
Once you've figured out what will be sent between the two players (manager, workers), how will the manager finish the communication with the workers? Define this part of the protocol.
Challenge 4 |
Now that you have a fairly complete protocol, write a skeleton MPI program that will support 1 manager and several workers, and where the protocol will be similar (although on a smaller scale) to the protocol you've defined above.
Test the program and run it.
Make sure it works if you have less data available than the number of workers created. In other words, make sure that if there are more workers than needed given the work, the manager will still be able to tell them to shut down at the right time.
Solution Protocol
Solution Source
Compile and Run
mpicc -o walkGetSizeAddMySql_Parallel_Nnodes $(mysql_config --cflags) \
walkGetSizeAddMySql_Parallel_Nnodes.c $(mysql_config --libs)
time mpirun -np 9 ./walkGetSizeAddMySql_Parallel_Nnodes \
/media/dominique/3TB/mediawiki/images/wikipedia/en/ \
0/05 25
Timing
Timing the effect of the parallelism provided by MPI on an 8-core AMD 3.0 GHz processor yields this data:
| # workers | block size (# files/block) |
real time | user time |
|---|---|---|---|
|
1 |
512 |
1m3.438s |
0m59.860s |
|
2 |
256 |
0m32.077s |
0m31.064s |
|
4 |
128 |
0m17.145s |
0m17.012s |
|
8 |
64 |
0m7.781s |
0m8.256s |
|
16 |
32 |
0m5.022s |
0m7.176s |
Note that there is speedup as the number of workers increase, with a definite lower return on parallelism as soon as we pass 8 workers (which makes sense as we have only 8 cores).
The bash script used to generate the data is shown below:
#! /bin/bash
noWorkers=( 1 2 4 8 16 )
blockSize=( 512 256 128 64 32 )
dim=${#noWorkers[*]}
lastIndex=$(( $dim - 1 ))
for i in $(seq 0 $lastIndex) ; do
echo "$i noWorkers = ${noWorkers[i]} blockSize = ${blockSize[i]}"
n=${noWorkers[i]}
noNodes=$(( $n + 1 ))
time mpirun -np $noNodes \
./walkGetSizeAddMySql_Parallel_Nnodes /media/dominique/3TB/mediawiki/images/wikipedia/en/ \
0/05 ${blockSize[i]}
done
